diff --git a/configs/calibration_benchmark.yaml b/configs/calibration_benchmark.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..830e73f30e4ad2a2fd3e0b62f3dcaa7baebe5471
--- /dev/null
+++ b/configs/calibration_benchmark.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - machine: aws
+  - model: default
+  - dataset: default
+  - _self_
+
+output_dir: ${hydra:run.dir}
+root_data_dir: ${machine.root_data_dir}
+mapanything_dataset_metadata_dir: ${machine.mapanything_dataset_metadata_dir}
+root_pretrained_checkpoints_dir: ${machine.root_pretrained_checkpoints_dir}
+root_experiments_dir: ${machine.root_experiments_dir}
+root_uniception_pretrained_checkpoints_dir: ${machine.root_uniception_pretrained_checkpoints_dir}
+
+### Benchmarking args
+seed: 0
+# Disable CUDNN Benchmark (Disable for variable resolution & number of view training)
+disable_cudnn_benchmark: true
+# Batch size for inference (Metrics are computed per multi-view set and averaged, not per batch of multi-view sets)
+batch_size: 20
+# Use mixed precision for inference
+amp: 1
+# Floating point type to use for mixed precision
+amp_dtype: "bf16"
diff --git a/configs/dataset/ase_wai/default.yaml b/configs/dataset/ase_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/ase_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/ase_wai/train/default.yaml b/configs/dataset/ase_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..5f900e2b75ee2acb9256253bc753f3c355d61335
--- /dev/null
+++ b/configs/dataset/ase_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ASEWAI(
+    split='${dataset.ase_wai.train.split}',
+    resolution=${dataset.ase_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.ase_wai.train.principal_point_centered},
+    aug_crop=${dataset.ase_wai.train.aug_crop},
+    transform='${dataset.ase_wai.train.transform}',
+    data_norm_type='${dataset.ase_wai.train.data_norm_type}',
+    ROOT='${dataset.ase_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.ase_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.ase_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.ase_wai.train.variable_num_views},
+    num_views=${dataset.ase_wai.train.num_views},
+    covisibility_thres=${dataset.ase_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/ase
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/ase_wai/val/default.yaml b/configs/dataset/ase_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..55457816cb985e4a1a4d05a3826fffcfe10878dc
--- /dev/null
+++ b/configs/dataset/ase_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ASEWAI(
+    split='${dataset.ase_wai.val.split}',
+    resolution=${dataset.ase_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.ase_wai.val.principal_point_centered},
+    seed=${dataset.ase_wai.val.seed},
+    transform='${dataset.ase_wai.val.transform}',
+    data_norm_type='${dataset.ase_wai.val.data_norm_type}',
+    ROOT='${dataset.ase_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.ase_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.ase_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.ase_wai.val.variable_num_views},
+    num_views=${dataset.ase_wai.val.num_views},
+    covisibility_thres=${dataset.ase_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_ase}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/ase
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/benchmark_512_eth3d_snpp_tav2.yaml b/configs/dataset/benchmark_512_eth3d_snpp_tav2.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..99c2db868664c48c66dbe46bc7907f7fbfd00f95
--- /dev/null
+++ b/configs/dataset/benchmark_512_eth3d_snpp_tav2.yaml
@@ -0,0 +1,20 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 2
+
+# Test Resolution
+resolution_test_eth3d: ${dataset.resolution_options.512_1_52_ar}
+resolution_test_scannetpp: ${dataset.resolution_options.512_1_52_ar}
+resolution_test_tav2_wb: ${dataset.resolution_options.512_1_00_ar}
+
+# Test Set
+# Sample 10 multi-view sets from each scene
+# ETH3D: 13 scenes
+# ScanNet++V2: 30 scenes
+# TartanAirV2-WB: 5 scenes
+test_dataset:
+  "+ 130 @ ${dataset.eth3d_wai.test.dataset_str}
+  + 300 @ ${dataset.scannetpp_wai.test.dataset_str}
+  + 50 @ ${dataset.tav2_wb_wai.test.dataset_str}"
diff --git a/configs/dataset/benchmark_518_eth3d_snpp_tav2.yaml b/configs/dataset/benchmark_518_eth3d_snpp_tav2.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4af1b75495bca7d575a56d621f8e1dc529d30170
--- /dev/null
+++ b/configs/dataset/benchmark_518_eth3d_snpp_tav2.yaml
@@ -0,0 +1,20 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 2
+
+# Test Resolution
+resolution_test_eth3d: ${dataset.resolution_options.518_1_52_ar}
+resolution_test_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_test_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+
+# Test Set
+# Sample 10 multi-view sets from each scene
+# ETH3D: 13 scenes
+# ScanNet++V2: 30 scenes
+# TartanAirV2-WB: 5 scenes
+test_dataset:
+  "+ 130 @ ${dataset.eth3d_wai.test.dataset_str}
+  + 300 @ ${dataset.scannetpp_wai.test.dataset_str}
+  + 50 @ ${dataset.tav2_wb_wai.test.dataset_str}"
diff --git a/configs/dataset/benchmark_sv_calib_518_many_ar_eth3d_snpp_tav2.yaml b/configs/dataset/benchmark_sv_calib_518_many_ar_eth3d_snpp_tav2.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..67da8ece5612aa1cd46dd0ae93aa80cc0d8ff1b7
--- /dev/null
+++ b/configs/dataset/benchmark_sv_calib_518_many_ar_eth3d_snpp_tav2.yaml
@@ -0,0 +1,20 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 1
+
+# Test Resolution
+resolution_test_eth3d: ${dataset.resolution_options.518_many_ar}
+resolution_test_scannetpp: ${dataset.resolution_options.518_many_ar}
+resolution_test_tav2_wb: ${dataset.resolution_options.518_many_ar}
+
+# Test Set
+# Sample 20 frames from each scene
+# ETH3D: 13 scenes
+# ScanNet++V2: 30 scenes
+# TartanAirV2-WB: 5 scenes
+test_dataset:
+  "+ 260 @ ${dataset.eth3d_wai.test.dataset_str}
+  + 600 @ ${dataset.scannetpp_wai.test.dataset_str}
+  + 100 @ ${dataset.tav2_wb_wai.test.dataset_str}"
diff --git a/configs/dataset/blendedmvs_wai/default.yaml b/configs/dataset/blendedmvs_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/blendedmvs_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/blendedmvs_wai/train/default.yaml b/configs/dataset/blendedmvs_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..63130dc49c81faf80aaf06c8ac545dac818b7746
--- /dev/null
+++ b/configs/dataset/blendedmvs_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "BlendedMVSWAI(
+    split='${dataset.blendedmvs_wai.train.split}',
+    resolution=${dataset.blendedmvs_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.blendedmvs_wai.train.principal_point_centered},
+    aug_crop=${dataset.blendedmvs_wai.train.aug_crop},
+    transform='${dataset.blendedmvs_wai.train.transform}',
+    data_norm_type='${dataset.blendedmvs_wai.train.data_norm_type}',
+    ROOT='${dataset.blendedmvs_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.blendedmvs_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.blendedmvs_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.blendedmvs_wai.train.variable_num_views},
+    num_views=${dataset.blendedmvs_wai.train.num_views},
+    covisibility_thres=${dataset.blendedmvs_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/blendedmvs
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/blendedmvs_wai/val/default.yaml b/configs/dataset/blendedmvs_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..9266a43316b8e278ce805478ea93d9fce585ec0a
--- /dev/null
+++ b/configs/dataset/blendedmvs_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "BlendedMVSWAI(
+    split='${dataset.blendedmvs_wai.val.split}',
+    resolution=${dataset.blendedmvs_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.blendedmvs_wai.val.principal_point_centered},
+    seed=${dataset.blendedmvs_wai.val.seed},
+    transform='${dataset.blendedmvs_wai.val.transform}',
+    data_norm_type='${dataset.blendedmvs_wai.val.data_norm_type}',
+    ROOT='${dataset.blendedmvs_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.blendedmvs_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.blendedmvs_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.blendedmvs_wai.val.variable_num_views},
+    num_views=${dataset.blendedmvs_wai.val.num_views},
+    covisibility_thres=${dataset.blendedmvs_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_blendedmvs}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/blendedmvs
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/bmvs_518_many_ar_48ipg_8g.yaml b/configs/dataset/bmvs_518_many_ar_48ipg_8g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..8aae5ec4353558a5c9893f55d3df4f0fc82fd939
--- /dev/null
+++ b/configs/dataset/bmvs_518_many_ar_48ipg_8g.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+
+# Training Set
+train_dataset:
+  "+ 140_000 @ ${dataset.blendedmvs_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}"
diff --git a/configs/dataset/default.yaml b/configs/dataset/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..828c780302408299b435f35fe47478105a4725f5
--- /dev/null
+++ b/configs/dataset/default.yaml
@@ -0,0 +1,39 @@
+defaults:
+  - resolution_options: default
+  - ase_wai: default
+  - blendedmvs_wai: default
+  - dl3dv_wai: default
+  - dynamicreplica_wai: default
+  - eth3d_wai: default
+  - megadepth_wai: default
+  - mpsd_wai: default
+  - mvs_synth_wai: default
+  - paralleldomain4d_wai: default
+  - sailvos3d_wai: default
+  - scannetpp_wai: default
+  - spring_wai: default
+  - tav2_wb_wai: default
+  - unrealstereo4k_wai: default
+
+# Training Set, For example: BlendedMVS(split='train', resolution=(512, 384), transform=...)
+train_dataset: ???
+# Validation Set
+test_dataset: "[null]"
+# Number of workers for dataloader
+num_workers: 12
+# Default resolution for training
+resolution_train: ???
+# Default resolution for validation
+resolution_val: ???
+# Number of views parameter for multi-view datasets
+num_views: 2
+# Use a centered principal point for all images
+principal_point_centered: false
+# Default config for multi-view datasets
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+val:
+  variable_num_views: false
+test:
+  variable_num_views: false
diff --git a/configs/dataset/dl3dv_wai/default.yaml b/configs/dataset/dl3dv_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/dl3dv_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/dl3dv_wai/train/default.yaml b/configs/dataset/dl3dv_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1ebbdc50856eaebef61a016b0c37e8301dbc541f
--- /dev/null
+++ b/configs/dataset/dl3dv_wai/train/default.yaml
@@ -0,0 +1,28 @@
+dataset_str:
+  "DL3DVWAI(
+    split='${dataset.dl3dv_wai.train.split}',
+    resolution=${dataset.dl3dv_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.dl3dv_wai.train.principal_point_centered},
+    aug_crop=${dataset.dl3dv_wai.train.aug_crop},
+    transform='${dataset.dl3dv_wai.train.transform}',
+    data_norm_type='${dataset.dl3dv_wai.train.data_norm_type}',
+    ROOT='${dataset.dl3dv_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.dl3dv_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.dl3dv_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.dl3dv_wai.train.variable_num_views},
+    num_views=${dataset.dl3dv_wai.train.num_views},
+    covisibility_thres=${dataset.dl3dv_wai.train.covisibility_thres},
+    mvs_confidence_filter_thres=${dataset.dl3dv_wai.train.mvs_confidence_filter_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/dl3dv
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
+mvs_confidence_filter_thres: 0.25
diff --git a/configs/dataset/dl3dv_wai/val/default.yaml b/configs/dataset/dl3dv_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..53f1d7d8d2ab1bb7849aae819d8d50d9c462ad2f
--- /dev/null
+++ b/configs/dataset/dl3dv_wai/val/default.yaml
@@ -0,0 +1,28 @@
+dataset_str:
+  "DL3DVWAI(
+    split='${dataset.dl3dv_wai.val.split}',
+    resolution=${dataset.dl3dv_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.dl3dv_wai.val.principal_point_centered},
+    seed=${dataset.dl3dv_wai.val.seed},
+    transform='${dataset.dl3dv_wai.val.transform}',
+    data_norm_type='${dataset.dl3dv_wai.val.data_norm_type}',
+    ROOT='${dataset.dl3dv_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.dl3dv_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.dl3dv_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.dl3dv_wai.val.variable_num_views},
+    num_views=${dataset.dl3dv_wai.val.num_views},
+    covisibility_thres=${dataset.dl3dv_wai.val.covisibility_thres},
+    mvs_confidence_filter_thres=${dataset.dl3dv_wai.val.mvs_confidence_filter_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_dl3dv}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/dl3dv
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
+mvs_confidence_filter_thres: 0.25
diff --git a/configs/dataset/dynamicreplica_wai/default.yaml b/configs/dataset/dynamicreplica_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/dynamicreplica_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/dynamicreplica_wai/train/default.yaml b/configs/dataset/dynamicreplica_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..62efc9b7cc342a46d2e74ae53969daf240caad83
--- /dev/null
+++ b/configs/dataset/dynamicreplica_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "DynamicReplicaWAI(
+    split='${dataset.dynamicreplica_wai.train.split}',
+    resolution=${dataset.dynamicreplica_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.dynamicreplica_wai.train.principal_point_centered},
+    aug_crop=${dataset.dynamicreplica_wai.train.aug_crop},
+    transform='${dataset.dynamicreplica_wai.train.transform}',
+    data_norm_type='${dataset.dynamicreplica_wai.train.data_norm_type}',
+    ROOT='${dataset.dynamicreplica_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.dynamicreplica_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.dynamicreplica_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.dynamicreplica_wai.train.variable_num_views},
+    num_views=${dataset.dynamicreplica_wai.train.num_views},
+    covisibility_thres=${dataset.dynamicreplica_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/dynamicreplica
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/dynamicreplica_wai/val/default.yaml b/configs/dataset/dynamicreplica_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..64f74ac5c6bf48beeccb656b27a4a4c7b05ce490
--- /dev/null
+++ b/configs/dataset/dynamicreplica_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "DynamicReplicaWAI(
+    split='${dataset.dynamicreplica_wai.val.split}',
+    resolution=${dataset.dynamicreplica_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.dynamicreplica_wai.val.principal_point_centered},
+    seed=${dataset.dynamicreplica_wai.val.seed},
+    transform='${dataset.dynamicreplica_wai.val.transform}',
+    data_norm_type='${dataset.dynamicreplica_wai.val.data_norm_type}',
+    ROOT='${dataset.dynamicreplica_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.dynamicreplica_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.dynamicreplica_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.dynamicreplica_wai.val.variable_num_views},
+    num_views=${dataset.dynamicreplica_wai.val.num_views},
+    covisibility_thres=${dataset.dynamicreplica_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_dynamicreplica}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/dynamicreplica
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/eth3d_wai/default.yaml b/configs/dataset/eth3d_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1da6ae479fdbefc1598057293f4d5966e6aefd51
--- /dev/null
+++ b/configs/dataset/eth3d_wai/default.yaml
@@ -0,0 +1,2 @@
+defaults:
+  - test: default
diff --git a/configs/dataset/eth3d_wai/test/default.yaml b/configs/dataset/eth3d_wai/test/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..fca5b2b491e50ed84ba56e5e540afc9dd18302bd
--- /dev/null
+++ b/configs/dataset/eth3d_wai/test/default.yaml
@@ -0,0 +1,22 @@
+dataset_str:
+  "ETH3DWAI(
+    resolution=${dataset.eth3d_wai.test.dataset_resolution},
+    principal_point_centered=${dataset.eth3d_wai.test.principal_point_centered},
+    seed=${dataset.eth3d_wai.test.seed},
+    transform='${dataset.eth3d_wai.test.transform}',
+    data_norm_type='${dataset.eth3d_wai.test.data_norm_type}',
+    ROOT='${dataset.eth3d_wai.test.ROOT}',
+    dataset_metadata_dir='${dataset.eth3d_wai.test.dataset_metadata_dir}',
+    variable_num_views=${dataset.eth3d_wai.test.variable_num_views},
+    num_views=${dataset.eth3d_wai.test.num_views},
+    covisibility_thres=${dataset.eth3d_wai.test.covisibility_thres})"
+dataset_resolution: ${dataset.resolution_test_eth3d}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/eth3d
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+variable_num_views: ${dataset.test.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.025
diff --git a/configs/dataset/megadepth_wai/default.yaml b/configs/dataset/megadepth_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/megadepth_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/megadepth_wai/train/default.yaml b/configs/dataset/megadepth_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a635383222cc8f505c7de8d9d8720bb550c7ceba
--- /dev/null
+++ b/configs/dataset/megadepth_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MegaDepthWAI(
+    split='${dataset.megadepth_wai.train.split}',
+    resolution=${dataset.megadepth_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.megadepth_wai.train.principal_point_centered},
+    aug_crop=${dataset.megadepth_wai.train.aug_crop},
+    transform='${dataset.megadepth_wai.train.transform}',
+    data_norm_type='${dataset.megadepth_wai.train.data_norm_type}',
+    ROOT='${dataset.megadepth_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.megadepth_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.megadepth_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.megadepth_wai.train.variable_num_views},
+    num_views=${dataset.megadepth_wai.train.num_views},
+    covisibility_thres=${dataset.megadepth_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/megadepth
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/megadepth_wai/val/default.yaml b/configs/dataset/megadepth_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1d40ed64bd64ca081192ca8973afb8feba1c65b4
--- /dev/null
+++ b/configs/dataset/megadepth_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MegaDepthWAI(
+    split='${dataset.megadepth_wai.val.split}',
+    resolution=${dataset.megadepth_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.megadepth_wai.val.principal_point_centered},
+    seed=${dataset.megadepth_wai.val.seed},
+    transform='${dataset.megadepth_wai.val.transform}',
+    data_norm_type='${dataset.megadepth_wai.val.data_norm_type}',
+    ROOT='${dataset.megadepth_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.megadepth_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.megadepth_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.megadepth_wai.val.variable_num_views},
+    num_views=${dataset.megadepth_wai.val.num_views},
+    covisibility_thres=${dataset.megadepth_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_megadepth}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/megadepth
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/megatrain_13d_512_many_ar_24ipg_16g.yaml b/configs/dataset/megatrain_13d_512_many_ar_24ipg_16g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c4cccb89d934132b1790b4f468d6d80c56d6827a
--- /dev/null
+++ b/configs/dataset/megatrain_13d_512_many_ar_24ipg_16g.yaml
@@ -0,0 +1,59 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.512_many_ar}
+
+# Validation Resolution
+resolution_val_ase: ${dataset.resolution_options.512_1_00_ar}
+resolution_val_blendedmvs: ${dataset.resolution_options.512_1_33_ar}
+resolution_val_dl3dv: ${dataset.resolution_options.512_1_77_ar}
+resolution_val_dynamicreplica: ${dataset.resolution_options.512_1_77_ar}
+resolution_val_megadepth: ${dataset.resolution_options.512_1_52_ar}
+resolution_val_mpsd: ${dataset.resolution_options.512_1_77_ar}
+resolution_val_mvs_synth: ${dataset.resolution_options.512_1_77_ar}
+resolution_val_paralleldomain4d: ${dataset.resolution_options.512_1_33_ar}
+resolution_val_sailvos3d: ${dataset.resolution_options.512_1_52_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.512_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.512_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.512_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.512_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 52_500 @ ${dataset.ase_wai.train.dataset_str}
+  + 52_500 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 52_500 @ ${dataset.dl3dv_wai.train.dataset_str}
+  + 40_000 @ ${dataset.dynamicreplica_wai.train.dataset_str}
+  + 52_500 @ ${dataset.megadepth_wai.train.dataset_str}
+  + 52_500 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 52_500 @ ${dataset.mvs_synth_wai.train.dataset_str}
+  + 52_500 @ ${dataset.paralleldomain4d_wai.train.dataset_str}
+  + 52_500 @ ${dataset.sailvos3d_wai.train.dataset_str}
+  + 52_500 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 2_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 52_500 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 5_500 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.ase_wai.val.dataset_str}
+  + 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dl3dv_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dynamicreplica_wai.val.dataset_str}
+  + 4_000 @ ${dataset.megadepth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mvs_synth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.paralleldomain4d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.sailvos3d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_13d_518_many_ar_24ipg_16g.yaml b/configs/dataset/megatrain_13d_518_many_ar_24ipg_16g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1c65bc4ebe5a8b22c79bfdf0e67fdcc95e068d75
--- /dev/null
+++ b/configs/dataset/megatrain_13d_518_many_ar_24ipg_16g.yaml
@@ -0,0 +1,59 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_ase: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_dl3dv: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_dynamicreplica: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_megadepth: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_mvs_synth: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_paralleldomain4d: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_sailvos3d: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 52_500 @ ${dataset.ase_wai.train.dataset_str}
+  + 52_500 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 52_500 @ ${dataset.dl3dv_wai.train.dataset_str}
+  + 40_000 @ ${dataset.dynamicreplica_wai.train.dataset_str}
+  + 52_500 @ ${dataset.megadepth_wai.train.dataset_str}
+  + 52_500 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 52_500 @ ${dataset.mvs_synth_wai.train.dataset_str}
+  + 52_500 @ ${dataset.paralleldomain4d_wai.train.dataset_str}
+  + 52_500 @ ${dataset.sailvos3d_wai.train.dataset_str}
+  + 52_500 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 2_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 52_500 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 5_500 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.ase_wai.val.dataset_str}
+  + 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dl3dv_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dynamicreplica_wai.val.dataset_str}
+  + 4_000 @ ${dataset.megadepth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mvs_synth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.paralleldomain4d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.sailvos3d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_13d_518_many_ar_24ipg_8g.yaml b/configs/dataset/megatrain_13d_518_many_ar_24ipg_8g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..10a5c74b6a6c1fd8ccdcdda090f9b4eb74fe6e9f
--- /dev/null
+++ b/configs/dataset/megatrain_13d_518_many_ar_24ipg_8g.yaml
@@ -0,0 +1,59 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_ase: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_dl3dv: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_dynamicreplica: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_megadepth: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_mvs_synth: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_paralleldomain4d: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_sailvos3d: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 26_250 @ ${dataset.ase_wai.train.dataset_str}
+  + 26_250 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 26_250 @ ${dataset.dl3dv_wai.train.dataset_str}
+  + 20_000 @ ${dataset.dynamicreplica_wai.train.dataset_str}
+  + 26_250 @ ${dataset.megadepth_wai.train.dataset_str}
+  + 26_250 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 26_250 @ ${dataset.mvs_synth_wai.train.dataset_str}
+  + 26_250 @ ${dataset.paralleldomain4d_wai.train.dataset_str}
+  + 26_250 @ ${dataset.sailvos3d_wai.train.dataset_str}
+  + 26_250 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 1_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 26_250 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 2_750 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.ase_wai.val.dataset_str}
+  + 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dl3dv_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dynamicreplica_wai.val.dataset_str}
+  + 4_000 @ ${dataset.megadepth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mvs_synth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.paralleldomain4d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.sailvos3d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_13d_518_many_ar_48ipg_64g.yaml b/configs/dataset/megatrain_13d_518_many_ar_48ipg_64g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d916f2a738e4cc0a9fc9c2df673509066b89581d
--- /dev/null
+++ b/configs/dataset/megatrain_13d_518_many_ar_48ipg_64g.yaml
@@ -0,0 +1,59 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_ase: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_dl3dv: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_dynamicreplica: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_megadepth: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_mvs_synth: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_paralleldomain4d: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_sailvos3d: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 420_000 @ ${dataset.ase_wai.train.dataset_str}
+  + 420_000 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 420_000 @ ${dataset.dl3dv_wai.train.dataset_str}
+  + 320_000 @ ${dataset.dynamicreplica_wai.train.dataset_str}
+  + 420_000 @ ${dataset.megadepth_wai.train.dataset_str}
+  + 420_000 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 420_000 @ ${dataset.mvs_synth_wai.train.dataset_str}
+  + 420_000 @ ${dataset.paralleldomain4d_wai.train.dataset_str}
+  + 420_000 @ ${dataset.sailvos3d_wai.train.dataset_str}
+  + 420_000 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 16_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 420_000 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 44_000 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.ase_wai.val.dataset_str}
+  + 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dl3dv_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dynamicreplica_wai.val.dataset_str}
+  + 4_000 @ ${dataset.megadepth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mvs_synth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.paralleldomain4d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.sailvos3d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_13d_518_many_ar_48ipg_8g_mono.yaml b/configs/dataset/megatrain_13d_518_many_ar_48ipg_8g_mono.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b9393fc3a45e1dc8410d6a2e7899c0dd42d27343
--- /dev/null
+++ b/configs/dataset/megatrain_13d_518_many_ar_48ipg_8g_mono.yaml
@@ -0,0 +1,59 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 1
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_ase: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_dl3dv: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_dynamicreplica: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_megadepth: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_mvs_synth: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_paralleldomain4d: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_sailvos3d: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 105_000 @ ${dataset.ase_wai.train.dataset_str}
+  + 105_000 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 105_000 @ ${dataset.dl3dv_wai.train.dataset_str}
+  + 80_000 @ ${dataset.dynamicreplica_wai.train.dataset_str}
+  + 105_000 @ ${dataset.megadepth_wai.train.dataset_str}
+  + 105_000 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 105_000 @ ${dataset.mvs_synth_wai.train.dataset_str}
+  + 105_000 @ ${dataset.paralleldomain4d_wai.train.dataset_str}
+  + 105_000 @ ${dataset.sailvos3d_wai.train.dataset_str}
+  + 105_000 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 4_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 105_000 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 11_000 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.ase_wai.val.dataset_str}
+  + 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dl3dv_wai.val.dataset_str}
+  + 4_000 @ ${dataset.dynamicreplica_wai.val.dataset_str}
+  + 4_000 @ ${dataset.megadepth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mvs_synth_wai.val.dataset_str}
+  + 4_000 @ ${dataset.paralleldomain4d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.sailvos3d_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_6d_518_many_ar_48ipg_64g.yaml b/configs/dataset/megatrain_6d_518_many_ar_48ipg_64g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..199d2637875dec96d629358b4ecbc549a72360f7
--- /dev/null
+++ b/configs/dataset/megatrain_6d_518_many_ar_48ipg_64g.yaml
@@ -0,0 +1,38 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 1_120_000 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 1_120_000 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 1_120_000 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 44_000 @ ${dataset.spring_wai.train.dataset_str}
+  + 1_120_000 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 116_000 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/megatrain_6d_518_many_ar_48ipg_8g.yaml b/configs/dataset/megatrain_6d_518_many_ar_48ipg_8g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..e83c5f2faeac7c7c081ee9fea1e6c0867c2d90e2
--- /dev/null
+++ b/configs/dataset/megatrain_6d_518_many_ar_48ipg_8g.yaml
@@ -0,0 +1,38 @@
+defaults:
+  - default
+
+# Number of views parameter for the multi-view datasets
+num_views: 4
+
+train:
+  # If True, the number of views can vary from batch to batch. The maximum number of views is num_views and minimum is 2. (On by default for N-view training)
+  variable_num_views: true
+
+# Train Resolution
+resolution_train: ${dataset.resolution_options.518_many_ar}
+
+# Validation Resolution
+resolution_val_blendedmvs: ${dataset.resolution_options.518_1_33_ar}
+resolution_val_mpsd: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_scannetpp: ${dataset.resolution_options.518_1_52_ar}
+resolution_val_spring: ${dataset.resolution_options.518_1_77_ar}
+resolution_val_tav2_wb: ${dataset.resolution_options.518_1_00_ar}
+resolution_val_unrealstereo4k: ${dataset.resolution_options.518_1_77_ar}
+
+# Training Set
+train_dataset:
+  "+ 140_000 @ ${dataset.blendedmvs_wai.train.dataset_str}
+  + 140_000 @ ${dataset.mpsd_wai.train.dataset_str}
+  + 140_000 @ ${dataset.scannetpp_wai.train.dataset_str}
+  + 5_500 @ ${dataset.spring_wai.train.dataset_str}
+  + 140_000 @ ${dataset.tav2_wb_wai.train.dataset_str}
+  + 14_500 @ ${dataset.unrealstereo4k_wai.train.dataset_str}"
+
+# Validation Set
+test_dataset:
+  "+ 4_000 @ ${dataset.blendedmvs_wai.val.dataset_str}
+  + 4_000 @ ${dataset.mpsd_wai.val.dataset_str}
+  + 4_000 @ ${dataset.scannetpp_wai.val.dataset_str}
+  + 500 @ ${dataset.spring_wai.val.dataset_str}
+  + 4_000 @ ${dataset.tav2_wb_wai.val.dataset_str}
+  + 500 @ ${dataset.unrealstereo4k_wai.val.dataset_str}"
diff --git a/configs/dataset/mpsd_wai/default.yaml b/configs/dataset/mpsd_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/mpsd_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/mpsd_wai/train/default.yaml b/configs/dataset/mpsd_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..f865279aaeb52a9ae41dd1004fbc2d544facade3
--- /dev/null
+++ b/configs/dataset/mpsd_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MPSDWAI(
+    split='${dataset.mpsd_wai.train.split}',
+    resolution=${dataset.mpsd_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.mpsd_wai.train.principal_point_centered},
+    aug_crop=${dataset.mpsd_wai.train.aug_crop},
+    transform='${dataset.mpsd_wai.train.transform}',
+    data_norm_type='${dataset.mpsd_wai.train.data_norm_type}',
+    ROOT='${dataset.mpsd_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.mpsd_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.mpsd_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.mpsd_wai.train.variable_num_views},
+    num_views=${dataset.mpsd_wai.train.num_views},
+    covisibility_thres=${dataset.mpsd_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/mpsd
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.15
diff --git a/configs/dataset/mpsd_wai/val/default.yaml b/configs/dataset/mpsd_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..6878f29f779791bd59a5d7e6c9733d93c14ca2f1
--- /dev/null
+++ b/configs/dataset/mpsd_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MPSDWAI(
+    split='${dataset.mpsd_wai.val.split}',
+    resolution=${dataset.mpsd_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.mpsd_wai.val.principal_point_centered},
+    seed=${dataset.mpsd_wai.val.seed},
+    transform='${dataset.mpsd_wai.val.transform}',
+    data_norm_type='${dataset.mpsd_wai.val.data_norm_type}',
+    ROOT='${dataset.mpsd_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.mpsd_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.mpsd_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.mpsd_wai.val.variable_num_views},
+    num_views=${dataset.mpsd_wai.val.num_views},
+    covisibility_thres=${dataset.mpsd_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_mpsd}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/mpsd
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.15
diff --git a/configs/dataset/mvs_synth_wai/default.yaml b/configs/dataset/mvs_synth_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/mvs_synth_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/mvs_synth_wai/train/default.yaml b/configs/dataset/mvs_synth_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c44c9ac70103da4a220c7145611033b369a29ba7
--- /dev/null
+++ b/configs/dataset/mvs_synth_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MVSSynthWAI(
+    split='${dataset.mvs_synth_wai.train.split}',
+    resolution=${dataset.mvs_synth_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.mvs_synth_wai.train.principal_point_centered},
+    aug_crop=${dataset.mvs_synth_wai.train.aug_crop},
+    transform='${dataset.mvs_synth_wai.train.transform}',
+    data_norm_type='${dataset.mvs_synth_wai.train.data_norm_type}',
+    ROOT='${dataset.mvs_synth_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.mvs_synth_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.mvs_synth_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.mvs_synth_wai.train.variable_num_views},
+    num_views=${dataset.mvs_synth_wai.train.num_views},
+    covisibility_thres=${dataset.mvs_synth_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/mvs_synth
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/mvs_synth_wai/val/default.yaml b/configs/dataset/mvs_synth_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..7cfb7f2629a4a8cab43aa2105f1988b0f87bf421
--- /dev/null
+++ b/configs/dataset/mvs_synth_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "MVSSynthWAI(
+    split='${dataset.mvs_synth_wai.val.split}',
+    resolution=${dataset.mvs_synth_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.mvs_synth_wai.val.principal_point_centered},
+    seed=${dataset.mvs_synth_wai.val.seed},
+    transform='${dataset.mvs_synth_wai.val.transform}',
+    data_norm_type='${dataset.mvs_synth_wai.val.data_norm_type}',
+    ROOT='${dataset.mvs_synth_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.mvs_synth_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.mvs_synth_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.mvs_synth_wai.val.variable_num_views},
+    num_views=${dataset.mvs_synth_wai.val.num_views},
+    covisibility_thres=${dataset.mvs_synth_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_mvs_synth}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/mvs_synth
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/paralleldomain4d_wai/default.yaml b/configs/dataset/paralleldomain4d_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/paralleldomain4d_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/paralleldomain4d_wai/train/default.yaml b/configs/dataset/paralleldomain4d_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..58b07d8d2cf00d71f567c425c7708119aa9fd985
--- /dev/null
+++ b/configs/dataset/paralleldomain4d_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ParallelDomain4DWAI(
+    split='${dataset.paralleldomain4d_wai.train.split}',
+    resolution=${dataset.paralleldomain4d_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.paralleldomain4d_wai.train.principal_point_centered},
+    aug_crop=${dataset.paralleldomain4d_wai.train.aug_crop},
+    transform='${dataset.paralleldomain4d_wai.train.transform}',
+    data_norm_type='${dataset.paralleldomain4d_wai.train.data_norm_type}',
+    ROOT='${dataset.paralleldomain4d_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.paralleldomain4d_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.paralleldomain4d_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.paralleldomain4d_wai.train.variable_num_views},
+    num_views=${dataset.paralleldomain4d_wai.train.num_views},
+    covisibility_thres=${dataset.paralleldomain4d_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/paralleldomain4d
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/paralleldomain4d_wai/val/default.yaml b/configs/dataset/paralleldomain4d_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..556d8cc57280fb786c5574cb77b8f46f426bd186
--- /dev/null
+++ b/configs/dataset/paralleldomain4d_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ParallelDomain4DWAI(
+    split='${dataset.paralleldomain4d_wai.val.split}',
+    resolution=${dataset.paralleldomain4d_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.paralleldomain4d_wai.val.principal_point_centered},
+    seed=${dataset.paralleldomain4d_wai.val.seed},
+    transform='${dataset.paralleldomain4d_wai.val.transform}',
+    data_norm_type='${dataset.paralleldomain4d_wai.val.data_norm_type}',
+    ROOT='${dataset.paralleldomain4d_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.paralleldomain4d_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.paralleldomain4d_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.paralleldomain4d_wai.val.variable_num_views},
+    num_views=${dataset.paralleldomain4d_wai.val.num_views},
+    covisibility_thres=${dataset.paralleldomain4d_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_paralleldomain4d}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/paralleldomain4d
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/resolution_options/default.yaml b/configs/dataset/resolution_options/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..942fe9efe3b1bb8eb92f99d3450ebbb25826c99f
--- /dev/null
+++ b/configs/dataset/resolution_options/default.yaml
@@ -0,0 +1,77 @@
+518_many_ar: '[(518, 518), (518, 392), (518, 336), (518, 294), (518, 252), (518, 168), (392, 518), (336, 518), (294, 518), (252, 518)]'
+518_many_landscape_ar: '[(518, 518), (518, 392), (518, 336), (518, 294), (518, 252), (518, 168)]'
+518_many_non_square_landscape_ar: '[(518, 392), (518, 336), (518, 294), (518, 252), (518, 168)]'
+518_0_50_ar: (252, 518) # 1:2
+518_0_56_ar: (294, 518) # 9:16
+518_0_66_ar: (336, 518) # 2:3
+518_0_75_ar: (392, 518) # 3:4
+518_1_00_ar: (518, 518) # 1:1
+518_1_33_ar: (518, 392) # 4:3
+518_1_52_ar: (518, 336) # 3:2
+518_1_77_ar: (518, 294) # 16:9
+518_2_00_ar: (518, 252) # 2:1
+518_3_20_ar: (518, 168) # 3.2:1
+512_many_ar: '[(512, 512), (512, 384), (512, 336), (512, 288), (512, 256), (512, 160), (384, 512), (336, 512), (288, 512), (256, 512)]'
+512_many_landscape_ar: '[(512, 512), (512, 384), (512, 336), (512, 288), (512, 256), (512, 160)]'
+512_many_non_square_landscape_ar: '[(512, 384), (512, 336), (512, 288), (512, 256), (512, 160)]'
+512_0_50_ar: (256, 512)
+512_0_56_ar: (288, 512)
+512_0_66_ar: (336, 512)
+512_0_75_ar: (384, 512)
+512_1_00_ar: (512, 512)
+512_1_33_ar: (512, 384)
+512_1_52_ar: (512, 336)
+512_1_77_ar: (512, 288)
+512_2_00_ar: (512, 256)
+512_3_20_ar: (512, 160)
+504_many_ar: '[(504, 504), (504, 378), (504, 322), (504, 280), (504, 238), (504, 154), (378, 504), (322, 504), (280, 504), (238, 504)]'
+504_many_landscape_ar: '[(504, 504), (504, 378), (504, 322), (504, 280), (504, 238), (504, 154)]'
+504_many_non_square_landscape_ar: '[(504, 378), (504, 322), (504, 280), (504, 238), (504, 154)]'
+504_0_50_ar: (238, 504)
+504_0_56_ar: (280, 504)
+504_0_66_ar: (322, 504)
+504_0_75_ar: (378, 504)
+504_1_00_ar: (504, 504)
+504_1_33_ar: (504, 378)
+504_1_52_ar: (504, 322)
+504_1_77_ar: (504, 280)
+504_2_00_ar: (504, 238)
+504_3_20_ar: (504, 154)
+448_many_ar: '[(448, 448), (448, 336), (448, 294), (448, 252), (448, 224), (448, 140), (336, 448), (294, 448), (252, 448), (224, 448)]'
+448_many_landscape_ar: '[(448, 448), (448, 336), (448, 294), (448, 252), (448, 224), (448, 140)]'
+448_many_non_square_landscape_ar: '[(448, 336), (448, 294), (448, 252), (448, 224), (448, 140)]'
+448_0_50_ar: (224, 448)
+448_0_56_ar: (252, 448)
+448_0_66_ar: (294, 448)
+448_0_75_ar: (336, 448)
+448_1_00_ar: (448, 448)
+448_1_33_ar: (448, 336)
+448_1_52_ar: (448, 294)
+448_1_77_ar: (448, 252)
+448_2_00_ar: (448, 224)
+448_3_20_ar: (448, 140)
+224_many_ar_14ps: '[(224, 224), (224, 168), (224, 154), (224, 126), (224, 112), (224, 70), (168, 224), (154, 224), (126, 224), (112, 224)]'
+224_many_landscape_ar_14ps: '[(224, 224), (224, 168), (224, 154), (224, 126), (224, 112), (224, 70)]'
+224_many_non_square_landscape_ar_14ps: '[(224, 168), (224, 154), (224, 126), (224, 112), (224, 70)]'
+224_0_50_ar_14ps: (112, 224)
+224_0_56_ar_14ps: (126, 224)
+224_0_66_ar_14ps: (154, 224)
+224_0_75_ar_14ps: (168, 224)
+224_1_00_ar: (224, 224)
+224_1_33_ar_14ps: (224, 168)
+224_1_52_ar_14ps: (224, 154)
+224_1_77_ar_14ps: (224, 126)
+224_2_00_ar_14ps: (224, 112)
+224_3_20_ar_14ps: (224, 70)
+224_many_ar_16ps: '[(224, 224), (224, 176), (224, 160), (224, 128), (224, 112), (224, 80), (176, 224), (160, 224), (128, 224), (112, 224)]'
+224_many_landscape_ar_16ps: '[(224, 224), (224, 176), (224, 160), (224, 128), (224, 112), (224, 80)]'
+224_many_non_square_landscape_ar_16ps: '[(224, 176), (224, 160), (224, 128), (224, 112), (224, 80)]'
+224_0_50_ar_16ps: (112, 224)
+224_0_56_ar_16ps: (128, 224)
+224_0_66_ar_16ps: (160, 224)
+224_0_75_ar_16ps: (176, 224)
+224_1_33_ar_16ps: (224, 176)
+224_1_52_ar_16ps: (224, 160)
+224_1_77_ar_16ps: (224, 128)
+224_2_00_ar_16ps: (224, 112)
+224_3_20_ar_16ps: (224, 80)
diff --git a/configs/dataset/sailvos3d_wai/default.yaml b/configs/dataset/sailvos3d_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/sailvos3d_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/sailvos3d_wai/train/default.yaml b/configs/dataset/sailvos3d_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..cbd1aec783eea443e35b7beb011fbce3006680cd
--- /dev/null
+++ b/configs/dataset/sailvos3d_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "SAILVOS3DWAI(
+    split='${dataset.sailvos3d_wai.train.split}',
+    resolution=${dataset.sailvos3d_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.sailvos3d_wai.train.principal_point_centered},
+    aug_crop=${dataset.sailvos3d_wai.train.aug_crop},
+    transform='${dataset.sailvos3d_wai.train.transform}',
+    data_norm_type='${dataset.sailvos3d_wai.train.data_norm_type}',
+    ROOT='${dataset.sailvos3d_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.sailvos3d_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.sailvos3d_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.sailvos3d_wai.train.variable_num_views},
+    num_views=${dataset.sailvos3d_wai.train.num_views},
+    covisibility_thres=${dataset.sailvos3d_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/sailvos3d
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/sailvos3d_wai/val/default.yaml b/configs/dataset/sailvos3d_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c1e24eca478a7722970d8bed9480dc3a8a009318
--- /dev/null
+++ b/configs/dataset/sailvos3d_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "SAILVOS3DWAI(
+    split='${dataset.sailvos3d_wai.val.split}',
+    resolution=${dataset.sailvos3d_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.sailvos3d_wai.val.principal_point_centered},
+    seed=${dataset.sailvos3d_wai.val.seed},
+    transform='${dataset.sailvos3d_wai.val.transform}',
+    data_norm_type='${dataset.sailvos3d_wai.val.data_norm_type}',
+    ROOT='${dataset.sailvos3d_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.sailvos3d_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.sailvos3d_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.sailvos3d_wai.val.variable_num_views},
+    num_views=${dataset.sailvos3d_wai.val.num_views},
+    covisibility_thres=${dataset.sailvos3d_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_sailvos3d}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/sailvos3d
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/scannetpp_wai/default.yaml b/configs/dataset/scannetpp_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4b0f64e569c04c066a5c120fba8b37d09472f1a0
--- /dev/null
+++ b/configs/dataset/scannetpp_wai/default.yaml
@@ -0,0 +1,4 @@
+defaults:
+  - train: default
+  - val: default
+  - test: default
diff --git a/configs/dataset/scannetpp_wai/test/default.yaml b/configs/dataset/scannetpp_wai/test/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..9aa9cb3aa115d018be59b287e24f2687d12918b2
--- /dev/null
+++ b/configs/dataset/scannetpp_wai/test/default.yaml
@@ -0,0 +1,24 @@
+dataset_str:
+  "ScanNetPPWAI(
+    split='${dataset.scannetpp_wai.test.split}',
+    resolution=${dataset.scannetpp_wai.test.dataset_resolution},
+    principal_point_centered=${dataset.scannetpp_wai.test.principal_point_centered},
+    seed=${dataset.scannetpp_wai.test.seed},
+    transform='${dataset.scannetpp_wai.test.transform}',
+    data_norm_type='${dataset.scannetpp_wai.test.data_norm_type}',
+    ROOT='${dataset.scannetpp_wai.test.ROOT}',
+    dataset_metadata_dir='${dataset.scannetpp_wai.test.dataset_metadata_dir}',
+    variable_num_views=${dataset.scannetpp_wai.test.variable_num_views},
+    num_views=${dataset.scannetpp_wai.test.num_views},
+    covisibility_thres=${dataset.scannetpp_wai.test.covisibility_thres})"
+split: 'test'
+dataset_resolution: ${dataset.resolution_test_scannetpp}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/scannetppv2
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+variable_num_views: ${dataset.test.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/scannetpp_wai/train/default.yaml b/configs/dataset/scannetpp_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1b3bc3df4b8d545face2117ef0f38b7ee4201ffc
--- /dev/null
+++ b/configs/dataset/scannetpp_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ScanNetPPWAI(
+    split='${dataset.scannetpp_wai.train.split}',
+    resolution=${dataset.scannetpp_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.scannetpp_wai.train.principal_point_centered},
+    aug_crop=${dataset.scannetpp_wai.train.aug_crop},
+    transform='${dataset.scannetpp_wai.train.transform}',
+    data_norm_type='${dataset.scannetpp_wai.train.data_norm_type}',
+    ROOT='${dataset.scannetpp_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.scannetpp_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.scannetpp_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.scannetpp_wai.train.variable_num_views},
+    num_views=${dataset.scannetpp_wai.train.num_views},
+    covisibility_thres=${dataset.scannetpp_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/scannetppv2
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/scannetpp_wai/val/default.yaml b/configs/dataset/scannetpp_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..8ffd3399537819c5fa924c459faf9dcd9afefd3d
--- /dev/null
+++ b/configs/dataset/scannetpp_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "ScanNetPPWAI(
+    split='${dataset.scannetpp_wai.val.split}',
+    resolution=${dataset.scannetpp_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.scannetpp_wai.val.principal_point_centered},
+    seed=${dataset.scannetpp_wai.val.seed},
+    transform='${dataset.scannetpp_wai.val.transform}',
+    data_norm_type='${dataset.scannetpp_wai.val.data_norm_type}',
+    ROOT='${dataset.scannetpp_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.scannetpp_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.scannetpp_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.scannetpp_wai.val.variable_num_views},
+    num_views=${dataset.scannetpp_wai.val.num_views},
+    covisibility_thres=${dataset.scannetpp_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_scannetpp}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/scannetppv2
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/spring_wai/default.yaml b/configs/dataset/spring_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/spring_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/spring_wai/train/default.yaml b/configs/dataset/spring_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..7a43893440933d40448af33bcdce02d321495261
--- /dev/null
+++ b/configs/dataset/spring_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "SpringWAI(
+    split='${dataset.spring_wai.train.split}',
+    resolution=${dataset.spring_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.spring_wai.train.principal_point_centered},
+    aug_crop=${dataset.spring_wai.train.aug_crop},
+    transform='${dataset.spring_wai.train.transform}',
+    data_norm_type='${dataset.spring_wai.train.data_norm_type}',
+    ROOT='${dataset.spring_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.spring_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.spring_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.spring_wai.train.variable_num_views},
+    num_views=${dataset.spring_wai.train.num_views},
+    covisibility_thres=${dataset.spring_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/spring
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/spring_wai/val/default.yaml b/configs/dataset/spring_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c285b319ae4eccba88e6abea7af721367fa03846
--- /dev/null
+++ b/configs/dataset/spring_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "SpringWAI(
+    split='${dataset.spring_wai.val.split}',
+    resolution=${dataset.spring_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.spring_wai.val.principal_point_centered},
+    seed=${dataset.spring_wai.val.seed},
+    transform='${dataset.spring_wai.val.transform}',
+    data_norm_type='${dataset.spring_wai.val.data_norm_type}',
+    ROOT='${dataset.spring_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.spring_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.spring_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.spring_wai.val.variable_num_views},
+    num_views=${dataset.spring_wai.val.num_views},
+    covisibility_thres=${dataset.spring_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_spring}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/spring
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/tav2_wb_wai/default.yaml b/configs/dataset/tav2_wb_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4b0f64e569c04c066a5c120fba8b37d09472f1a0
--- /dev/null
+++ b/configs/dataset/tav2_wb_wai/default.yaml
@@ -0,0 +1,4 @@
+defaults:
+  - train: default
+  - val: default
+  - test: default
diff --git a/configs/dataset/tav2_wb_wai/test/default.yaml b/configs/dataset/tav2_wb_wai/test/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..ab5dcfb90bd43c0e4a48b9c0af4390afb820c115
--- /dev/null
+++ b/configs/dataset/tav2_wb_wai/test/default.yaml
@@ -0,0 +1,24 @@
+dataset_str:
+  "TartanAirV2WBWAI(
+    split='${dataset.tav2_wb_wai.test.split}',
+    resolution=${dataset.tav2_wb_wai.test.dataset_resolution},
+    principal_point_centered=${dataset.tav2_wb_wai.test.principal_point_centered},
+    seed=${dataset.tav2_wb_wai.test.seed},
+    transform='${dataset.tav2_wb_wai.test.transform}',
+    data_norm_type='${dataset.tav2_wb_wai.test.data_norm_type}',
+    ROOT='${dataset.tav2_wb_wai.test.ROOT}',
+    dataset_metadata_dir='${dataset.tav2_wb_wai.test.dataset_metadata_dir}',
+    variable_num_views=${dataset.tav2_wb_wai.test.variable_num_views},
+    num_views=${dataset.tav2_wb_wai.test.num_views},
+    covisibility_thres=${dataset.tav2_wb_wai.test.covisibility_thres})"
+split: 'test'
+dataset_resolution: ${dataset.resolution_test_tav2_wb}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/tav2_wb
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+variable_num_views: ${dataset.test.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/tav2_wb_wai/train/default.yaml b/configs/dataset/tav2_wb_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..adff13bba769d76aba5ae5550887896312405f29
--- /dev/null
+++ b/configs/dataset/tav2_wb_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "TartanAirV2WBWAI(
+    split='${dataset.tav2_wb_wai.train.split}',
+    resolution=${dataset.tav2_wb_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.tav2_wb_wai.train.principal_point_centered},
+    aug_crop=${dataset.tav2_wb_wai.train.aug_crop},
+    transform='${dataset.tav2_wb_wai.train.transform}',
+    data_norm_type='${dataset.tav2_wb_wai.train.data_norm_type}',
+    ROOT='${dataset.tav2_wb_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.tav2_wb_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.tav2_wb_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.tav2_wb_wai.train.variable_num_views},
+    num_views=${dataset.tav2_wb_wai.train.num_views},
+    covisibility_thres=${dataset.tav2_wb_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/tav2_wb
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/tav2_wb_wai/val/default.yaml b/configs/dataset/tav2_wb_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b6b2cb5925ea9f911ddecbb391a24a241636da66
--- /dev/null
+++ b/configs/dataset/tav2_wb_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "TartanAirV2WBWAI(
+    split='${dataset.tav2_wb_wai.val.split}',
+    resolution=${dataset.tav2_wb_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.tav2_wb_wai.val.principal_point_centered},
+    seed=${dataset.tav2_wb_wai.val.seed},
+    transform='${dataset.tav2_wb_wai.val.transform}',
+    data_norm_type='${dataset.tav2_wb_wai.val.data_norm_type}',
+    ROOT='${dataset.tav2_wb_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.tav2_wb_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.tav2_wb_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.tav2_wb_wai.val.variable_num_views},
+    num_views=${dataset.tav2_wb_wai.val.num_views},
+    covisibility_thres=${dataset.tav2_wb_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_tav2_wb}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/tav2_wb
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/unrealstereo4k_wai/default.yaml b/configs/dataset/unrealstereo4k_wai/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..909f9d015c19d0e7a075102484f5c711aa1f3bf3
--- /dev/null
+++ b/configs/dataset/unrealstereo4k_wai/default.yaml
@@ -0,0 +1,3 @@
+defaults:
+  - train: default
+  - val: default
diff --git a/configs/dataset/unrealstereo4k_wai/train/default.yaml b/configs/dataset/unrealstereo4k_wai/train/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..947161e6ea2b766bddd2083417a55647df0a45ed
--- /dev/null
+++ b/configs/dataset/unrealstereo4k_wai/train/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "UnrealStereo4KWAI(
+    split='${dataset.unrealstereo4k_wai.train.split}',
+    resolution=${dataset.unrealstereo4k_wai.train.dataset_resolution},
+    principal_point_centered=${dataset.unrealstereo4k_wai.train.principal_point_centered},
+    aug_crop=${dataset.unrealstereo4k_wai.train.aug_crop},
+    transform='${dataset.unrealstereo4k_wai.train.transform}',
+    data_norm_type='${dataset.unrealstereo4k_wai.train.data_norm_type}',
+    ROOT='${dataset.unrealstereo4k_wai.train.ROOT}',
+    dataset_metadata_dir='${dataset.unrealstereo4k_wai.train.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.unrealstereo4k_wai.train.overfit_num_sets},
+    variable_num_views=${dataset.unrealstereo4k_wai.train.variable_num_views},
+    num_views=${dataset.unrealstereo4k_wai.train.num_views},
+    covisibility_thres=${dataset.unrealstereo4k_wai.train.covisibility_thres})"
+split: 'train'
+dataset_resolution: ${dataset.resolution_train}
+principal_point_centered: ${dataset.principal_point_centered}
+aug_crop: 16
+transform: 'colorjitter+grayscale+gaublur'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/unrealstereo4k
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.train.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dataset/unrealstereo4k_wai/val/default.yaml b/configs/dataset/unrealstereo4k_wai/val/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..acf919e5745e604943a0ab41e6d8438108c1b25a
--- /dev/null
+++ b/configs/dataset/unrealstereo4k_wai/val/default.yaml
@@ -0,0 +1,26 @@
+dataset_str:
+  "UnrealStereo4KWAI(
+    split='${dataset.unrealstereo4k_wai.val.split}',
+    resolution=${dataset.unrealstereo4k_wai.val.dataset_resolution},
+    principal_point_centered=${dataset.unrealstereo4k_wai.val.principal_point_centered},
+    seed=${dataset.unrealstereo4k_wai.val.seed},
+    transform='${dataset.unrealstereo4k_wai.val.transform}',
+    data_norm_type='${dataset.unrealstereo4k_wai.val.data_norm_type}',
+    ROOT='${dataset.unrealstereo4k_wai.val.ROOT}',
+    dataset_metadata_dir='${dataset.unrealstereo4k_wai.val.dataset_metadata_dir}',
+    overfit_num_sets=${dataset.unrealstereo4k_wai.val.overfit_num_sets},
+    variable_num_views=${dataset.unrealstereo4k_wai.val.variable_num_views},
+    num_views=${dataset.unrealstereo4k_wai.val.num_views},
+    covisibility_thres=${dataset.unrealstereo4k_wai.val.covisibility_thres})"
+split: 'val'
+dataset_resolution: ${dataset.resolution_val_unrealstereo4k}
+principal_point_centered: ${dataset.principal_point_centered}
+seed: 777
+transform: 'imgnorm'
+data_norm_type: ${model.data_norm_type}
+ROOT: ${root_data_dir}/unrealstereo4k
+dataset_metadata_dir: ${mapanything_dataset_metadata_dir}
+overfit_num_sets: null
+variable_num_views: ${dataset.val.variable_num_views}
+num_views: ${dataset.num_views}
+covisibility_thres: 0.25
diff --git a/configs/dense_n_view_benchmark.yaml b/configs/dense_n_view_benchmark.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..90a60868f0e6ba698f02c711d0b1e0a57b2175f9
--- /dev/null
+++ b/configs/dense_n_view_benchmark.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - machine: aws
+  - model: default
+  - dataset: default
+  - _self_
+
+output_dir: ${hydra:run.dir}
+root_data_dir: ${machine.root_data_dir}
+mapanything_dataset_metadata_dir: ${machine.mapanything_dataset_metadata_dir}
+root_pretrained_checkpoints_dir: ${machine.root_pretrained_checkpoints_dir}
+root_experiments_dir: ${machine.root_experiments_dir}
+root_uniception_pretrained_checkpoints_dir: ${machine.root_uniception_pretrained_checkpoints_dir}
+
+### Benchmarking args
+seed: 0
+# Disable CUDNN Benchmark (Disable for variable resolution & number of view training)
+disable_cudnn_benchmark: true
+# Batch size for inference (Metrics are computed per multi-view set and averaged, not per batch of multi-view sets)
+batch_size: 10
+# Use mixed precision for inference
+amp: 1
+# Floating point type to use for mixed precision
+amp_dtype: "bf16"
diff --git a/configs/distributed/default.yaml b/configs/distributed/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..0e54fe64b7bf8879fed89eac6d21e08a96769c64
--- /dev/null
+++ b/configs/distributed/default.yaml
@@ -0,0 +1,6 @@
+# Distributed Training Params
+# Number of distributed processes
+world_size: 1
+local_rank: -1
+# Url used to set up distributed training
+dist_url: 'env://'
diff --git a/configs/loss/conf_pm_mask_loss.yaml b/configs/loss/conf_pm_mask_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..6375cf1dcaf054f14a2f248ccfb38801a93391be
--- /dev/null
+++ b/configs/loss/conf_pm_mask_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfLoss(Regr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', loss_in_log=True), alpha=0.2) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(Regr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', flatten_across_image_only=True, loss_in_log=True), top_n_percent=5, apply_to_real_data_only=True) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/conf_pm_mask_scale_loss.yaml b/configs/loss/conf_pm_mask_scale_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..46f754258cabcf7f9c61c69b8e541196aca2aa2d
--- /dev/null
+++ b/configs/loss/conf_pm_mask_scale_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfLoss(PointsPlusScaleRegr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', flatten_across_image_only=True, loss_in_log=True), alpha=0.2) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(PointsPlusScaleRegr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', flatten_across_image_only=True, loss_in_log=True), top_n_percent=5, apply_to_real_data_only=True) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/default.yaml b/configs/loss/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..5db7d1b5a58dd7b08185e530b462b9d0f3b47f50
--- /dev/null
+++ b/configs/loss/default.yaml
@@ -0,0 +1,6 @@
+# Training Loss, For example: "ConfLoss(Regr3D(L21, norm_mode='avg_dis'), alpha=0.2)"
+train_criterion: ""
+# Validation Loss, For example:
+# "Regr3D_ScaleShiftInv(L21, norm_mode='?avg_dis', ambiguous_loss_value=0)" (DUSt3R)
+# "Regr3D(L21, norm_mode='?avg_dis', ambiguous_loss_value=2)" (MASt3R)
+test_criterion: ""
diff --git a/configs/loss/entangled_metric_loss.yaml b/configs/loss/entangled_metric_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..39178aaf8c27ee42cd690c0047b38bfe46476c07
--- /dev/null
+++ b/configs/loss/entangled_metric_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/moge2_loss.yaml b/configs/loss/moge2_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..31bcd338f4ba79e2daac4422a1ca3ddcdb0a6b4f
--- /dev/null
+++ b/configs/loss/moge2_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ExcludeTopNPercentPixelLoss(Regr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', loss_in_log=True, flatten_across_image_only=True), top_n_percent=5, apply_to_real_data_only=True) + 3.0 * NormalGMLoss(norm_mode='avg_dis', apply_normal_and_gm_loss_to_synthetic_data_only=True)"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(Regr3D(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='?avg_dis', loss_in_log=True, flatten_across_image_only=True), top_n_percent=5, apply_to_real_data_only=True) + 3.0 * NormalGMLoss(norm_mode='avg_dis', apply_normal_and_gm_loss_to_synthetic_data_only=True)"
diff --git a/configs/loss/no_depth_loss.yaml b/configs/loss/no_depth_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..ce1deee438d4c84fa6112b316c1f663680d6fd2c
--- /dev/null
+++ b/configs/loss/no_depth_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, depth_loss_weight=0.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, depth_loss_weight=0.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/no_log_scaling.yaml b/configs/loss/no_log_scaling.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..f30447ceb31c783a3c62ecb2666c1e7572a85e03
--- /dev/null
+++ b/configs/loss/no_log_scaling.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=False, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=False, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/no_points_loss.yaml b/configs/loss/no_points_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..6680702c3870e5b137d6042b636197907adb94c4
--- /dev/null
+++ b/configs/loss/no_points_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=False, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, cam_frame_points_loss_weight=0.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[1]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=False, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, cam_frame_points_loss_weight=0.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[1]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/no_pose_loss.yaml b/configs/loss/no_pose_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b804c79999266f734b61602f8e8785c46c11d266
--- /dev/null
+++ b/configs/loss/no_pose_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, pose_quats_loss_weight=0.0, pose_trans_loss_weight=0.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, pose_quats_loss_weight=0.0, pose_trans_loss_weight=0.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/no_ray_dirs_loss.yaml b/configs/loss/no_ray_dirs_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a584054cf610cbb9ab2dbcd98878ed57b87b43b6
--- /dev/null
+++ b/configs/loss/no_ray_dirs_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, ray_directions_loss_weight=0.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0, ray_directions_loss_weight=0.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/no_robust_loss.yaml b/configs/loss/no_robust_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..efbcfd30f90f7d4568a3d28b9d7e99a93e1be069
--- /dev/null
+++ b/configs/loss/no_robust_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(L2Loss(), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=1.0, gm_loss_weight=1.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.1 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(L2Loss(), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=1.0, gm_loss_weight=1.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.1 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/overall_disentangled_loss.yaml b/configs/loss/overall_disentangled_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..f18662efe888e8f66a56c3297270fa9ee8cc2691
--- /dev/null
+++ b/configs/loss/overall_disentangled_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfLoss(DisentangledFactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), alpha=0.2, loss_set_indices=[0]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(DisentangledFactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/overall_loss.yaml b/configs/loss/overall_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c25db5cbb2b23667b1302d0b14ae497f684a5b13
--- /dev/null
+++ b/configs/loss/overall_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/overall_loss_weigh_pm_higher.yaml b/configs/loss/overall_loss_weigh_pm_higher.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c256b23d4f148b12d8e920fcde40cfa404754b5c
--- /dev/null
+++ b/configs/loss/overall_loss_weigh_pm_higher.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, cam_frame_points_loss_weight=0.1, depth_loss_weight=0.1, ray_directions_loss_weight=0.1, pose_quats_loss_weight=0.1, pose_trans_loss_weight=0.1, scale_loss_weight=0.1, world_frame_points_loss_weight=1, normal_loss_weight=0.3, gm_loss_weight=0.3), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.03 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryScaleRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, cam_frame_points_loss_weight=0.1, depth_loss_weight=0.1, ray_directions_loss_weight=0.1, pose_quats_loss_weight=0.1, pose_trans_loss_weight=0.1, scale_loss_weight=0.1, world_frame_points_loss_weight=1, normal_loss_weight=0.3, gm_loss_weight=0.3), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.03 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/pi3_loss.yaml b/configs/loss/pi3_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..09f2345df7d1b2584b7497fb0d6ce9b31e03d3f2
--- /dev/null
+++ b/configs/loss/pi3_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_z', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=True, convert_predictions_to_view0_frame=True, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2])"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_z', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=True, convert_predictions_to_view0_frame=True, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2])"
diff --git a/configs/loss/up_to_scale_loss.yaml b/configs/loss/up_to_scale_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..50503a96b4275a3ca2d301da0c37aab161a94f7e
--- /dev/null
+++ b/configs/loss/up_to_scale_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_along_ray', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2]) + 0.3 * NonAmbiguousMaskLoss(BCELoss())"
diff --git a/configs/loss/vggt_loss.yaml b/configs/loss/vggt_loss.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..213571b18798ed7bc0256db1916a8c3058e12009
--- /dev/null
+++ b/configs/loss/vggt_loss.yaml
@@ -0,0 +1,4 @@
+# Training Loss
+train_criterion: "ConfAndExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_z', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), conf_alpha=0.2, top_n_percent=5, apply_to_real_data_only=True, conf_loss_set_indices=[0], exclude_loss_set_indices=[1, 2])"
+# Validation Loss
+test_criterion: "ExcludeTopNPercentPixelLoss(FactoredGeometryRegr3DPlusNormalGMLoss(RobustRegressionLoss(alpha=0.5, scaling_c=0.05), norm_mode='avg_dis', depth_type_for_loss='depth_z', loss_in_log=True, flatten_across_image_only=True, compute_pairwise_relative_pose_loss=False, compute_world_frame_points_loss=True, apply_normal_and_gm_loss_to_synthetic_data_only=True, normal_loss_weight=3.0, gm_loss_weight=3.0), top_n_percent=5, apply_to_real_data_only=True, loss_set_indices=[0, 1, 2])"
diff --git a/configs/machine/aws.yaml b/configs/machine/aws.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..71f0d31bc676008dfdc2fde4231af93aa00d0869
--- /dev/null
+++ b/configs/machine/aws.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - default
+
+# Root directory containing all datasets
+root_data_dir: "/fsx/xrtech/data"
+# Dataset metadata directory
+mapanything_dataset_metadata_dir: "/fsx/nkeetha/mapanything_dataset_metadata"
+# Root directory containing pretrained checkpoints for custom models
+root_pretrained_checkpoints_dir: "/fsx/nkeetha/mapanything_checkpoints"
+# Root directory to log experiments
+root_experiments_dir: "/fsx/nkeetha/experiments"
+# Root directory containing UniCeption pretrained checkpoints
+root_uniception_pretrained_checkpoints_dir: "/fsx/nkeetha/uniception_checkpoints"
+# Root directory containing external benchmark data
+external_benchmark_data_root_data_dir: "/fsx/xrtech/external_benchmark_data/rmvd_mvs_benchmark/rmvd_test_data"
diff --git a/configs/machine/default.yaml b/configs/machine/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3c9852a356536a333046429588266258f9328bbe
--- /dev/null
+++ b/configs/machine/default.yaml
@@ -0,0 +1,12 @@
+# Root directory containing all datasets
+root_data_dir: ???
+# Dataset metadata directory
+mapanything_dataset_metadata_dir: ???
+# Root directory containing pretrained checkpoints for custom models
+root_pretrained_checkpoints_dir: ???
+# Root directory to log experiments
+root_experiments_dir: ???
+# Root directory containing UniCeption pretrained checkpoints
+root_uniception_pretrained_checkpoints_dir: ???
+# Root directory containing external benchmark data
+external_benchmark_data_root_data_dir: ???
diff --git a/configs/machine/psc.yaml b/configs/machine/psc.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..342448e85bf7ddb0ee6bd97e9ef39c067d15cc34
--- /dev/null
+++ b/configs/machine/psc.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - default
+
+# Root directory containing all datasets
+root_data_dir: "/ocean/projects/cis220039p/shared/datasets"
+# Dataset metadata directory
+mapanything_dataset_metadata_dir: "/ocean/projects/cis220039p/shared/mapanything_dataset_metadata"
+# Root directory containing pretrained checkpoints for custom models
+root_pretrained_checkpoints_dir: "/jet/home/yzhang25/mapanything/checkpoints"
+# Root directory to log experiments
+root_experiments_dir: "/jet/home/yzhang25/mapanything/outputs"
+# Root directory containing UniCeption pretrained checkpoints
+root_uniception_pretrained_checkpoints_dir: "/ocean/projects/cis220039p/shared/uniception/checkpoints/"
+# Root directory containing external benchmark data
+external_benchmark_data_root_data_dir: "/jet/home/yzhang25/mapanything/benchmarking/rmvd_mvs_benchmark/rmvd_test_data"
diff --git a/configs/machine/xri_dgx.yaml b/configs/machine/xri_dgx.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..cb4ac9a8523e9a8de994d71f5071e19641b739f7
--- /dev/null
+++ b/configs/machine/xri_dgx.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - default
+
+# Root directory containing all datasets
+root_data_dir: "/mnt/xri_mapsresearch/data/nkeetha"
+# Dataset metadata directory
+mapanything_dataset_metadata_dir: "/mnt/xri_mapsresearch/data/nkeetha/mapanything_dataset_metadata"
+# Root directory containing pretrained checkpoints for custom models
+root_pretrained_checkpoints_dir: "/mnt/xri_mapsresearch/code/nkeetha/mapanything/checkpoints"
+# Root directory to log experiments
+root_experiments_dir: "/mnt/xri_mapsresearch/experiments/nkeetha"
+# Root directory containing UniCeption pretrained checkpoints
+root_uniception_pretrained_checkpoints_dir: "/mnt/xri_mapsresearch/code/nkeetha/mapanything/UniCeption/checkpoints"
+# Root directory containing external benchmark data
+external_benchmark_data_root_data_dir: "/mnt/xri_mapsresearch/data/nkeetha/rmvd_mvs_benchmark/rmvd_test_data"
diff --git a/configs/model/anycalib.yaml b/configs/model/anycalib.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..0e73f8a68fdc58d98d72af2121ebc21c52788a26
--- /dev/null
+++ b/configs/model/anycalib.yaml
@@ -0,0 +1,11 @@
+# String for model factory
+model_str: "anycalib"
+# Model config
+model_config:
+  name: "anycalib"
+# Image Normalization Type
+data_norm_type: "identity"
+# AnyCalib checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/default.yaml b/configs/model/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3e3864d11fbfd9c575df77598ba42a218df8857c
--- /dev/null
+++ b/configs/model/default.yaml
@@ -0,0 +1,16 @@
+# String for model factory (Options: "mapanything", "mapanything_ablations", "modular_dust3r", "vggt", "pi3")
+model_str: ???
+# Model config
+model_config:
+  # Path to pretrained model checkpoint
+  pretrained_checkpoint_path: null
+  # Load specific submodules from the checkpoint
+  load_specific_pretrained_submodules: False
+  # List of submodules to load from the checkpoint (if load_specific_pretrained_submodules is True)
+  specific_pretrained_submodules: []
+# Path of a starting checkpoint (to enable backward compatibility with original DUSt3R class)
+pretrained: null
+# Image normalization type
+data_norm_type: ???
+# Torch hub force reload
+torch_hub_force_reload: False
diff --git a/configs/model/dust3r.yaml b/configs/model/dust3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..e9eb502d305d96cad2620f947e31e15f5a292d5c
--- /dev/null
+++ b/configs/model/dust3r.yaml
@@ -0,0 +1,23 @@
+# String for model factory
+model_str: "dust3r"
+# Model config
+model_config:
+  name: "dust3r"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/DUSt3R_ViTLarge_BaseDecoder_512_dpt.pth"
+  # Scene graph for BA
+  scene_graph: "complete"
+  # Pairwise inference batch size
+  inference_batch_size: 32
+  # Global optim schedule
+  global_optim_schedule: "cosine"
+  # Global optim lr
+  global_optim_lr: 0.01
+  # Number of iterations for global optimization
+  global_optim_niter: 300
+# Image Normalization Type
+data_norm_type: "dust3r"
+# DUSt3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/encoder/croco_512.yaml b/configs/model/encoder/croco_512.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..af463f7c116f799d90234a3315d69145723fdf02
--- /dev/null
+++ b/configs/model/encoder/croco_512.yaml
@@ -0,0 +1,16 @@
+# UniCeption encoder string used for selecting encoder class (python3 -m uniception.models.encoders.list)
+encoder_str: "croco"
+# Name of the encoder
+name: "croco_512"
+# Data normalization type
+data_norm_type: "croco"
+# Patch embedding class
+patch_embed_cls: "PatchEmbedDust3R"
+# Image size
+img_size: [512, 512] # This parameter has no influence for PatchEmbedDust3R
+# Path to the pretrained encoder checkpoint
+pretrained_checkpoint_path: '${machine.root_uniception_pretrained_checkpoints_dir}/encoders/CroCo_Encoder_224.pth'
+# Override attributes in the pretrained checkpoint
+override_checkpoint_attributes: True
+# Flag to indicate whether model class uses torch hub
+uses_torch_hub: False
diff --git a/configs/model/encoder/croco_512_data_norm_dust3r.yaml b/configs/model/encoder/croco_512_data_norm_dust3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d6158ba02e357994a2599addee8f61efddf0c17d
--- /dev/null
+++ b/configs/model/encoder/croco_512_data_norm_dust3r.yaml
@@ -0,0 +1,16 @@
+# UniCeption encoder string used for selecting encoder class (python3 -m uniception.models.encoders.list)
+encoder_str: "croco"
+# Name of the encoder
+name: "croco_512_img_norm_dust3r"
+# Data normalization type
+data_norm_type: "dust3r"
+# Patch embedding class
+patch_embed_cls: "PatchEmbedDust3R"
+# Image size
+img_size: [512, 512] # This parameter has no influence for PatchEmbedDust3R
+# Path to the pretrained encoder checkpoint
+pretrained_checkpoint_path: '${machine.root_uniception_pretrained_checkpoints_dir}/encoders/CroCo_Encoder_224.pth'
+# Override attributes in the pretrained checkpoint
+override_checkpoint_attributes: True
+# Flag to indicate whether model class uses torch hub
+uses_torch_hub: False
diff --git a/configs/model/encoder/dinov2_large.yaml b/configs/model/encoder/dinov2_large.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..2e1aa0ee137a4dfcc529ec633989e892b344b675
--- /dev/null
+++ b/configs/model/encoder/dinov2_large.yaml
@@ -0,0 +1,14 @@
+# UniCeption encoder string used for selecting encoder class (python3 -m uniception.models.encoders.list)
+encoder_str: "dinov2"
+# Name of the encoder
+name: "dinov2_large"
+# Data normalization type
+data_norm_type: "dinov2"
+# ViT size
+size: "large"
+# Registers
+with_registers: False
+# Flag to indicate whether model class uses torch hub
+uses_torch_hub: True
+# Flag to indicate whether to use gradient checkpointing for encoder
+gradient_checkpointing: False
diff --git a/configs/model/encoder/radio_v2_5_large.yaml b/configs/model/encoder/radio_v2_5_large.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..6236761e280b9193001e035bc703efd9869bf9a8
--- /dev/null
+++ b/configs/model/encoder/radio_v2_5_large.yaml
@@ -0,0 +1,10 @@
+# UniCeption encoder string used for selecting encoder class (python3 -m uniception.models.encoders.list)
+encoder_str: "radio"
+# Name of the encoder
+name: "radio_v2.5-large"
+# Data normalization type
+data_norm_type: "radio"
+# Model version
+model_version: "radio_v2.5-l"
+# Flag to indicate whether model class uses torch hub
+uses_torch_hub: True
diff --git a/configs/model/info_sharing/aat_ifr_24_layers.yaml b/configs/model/info_sharing/aat_ifr_24_layers.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4aeabafc404a09a64769064a874408e98818fa2e
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_24_layers.yaml
@@ -0,0 +1,22 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_24_layers_ifr"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/aat_ifr_24_layers_escaling.yaml b/configs/model/info_sharing/aat_ifr_24_layers_escaling.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..60f461af9f439dfd86cf3d81f565fcd65612f2cc
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_24_layers_escaling.yaml
@@ -0,0 +1,24 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_24_layers_ifr"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: True
+  # Scale Entropy in Attention
+  use_entropy_scaling: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/aat_ifr_24_layers_no_ref_view.yaml b/configs/model/info_sharing/aat_ifr_24_layers_no_ref_view.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..42edf9dfcd2c1d5986147f4db2444e4845bcd4f9
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_24_layers_no_ref_view.yaml
@@ -0,0 +1,22 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_24_layers_ifr_no_ref_view"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: False
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/aat_ifr_24_layers_w_view_pe.yaml b/configs/model/info_sharing/aat_ifr_24_layers_w_view_pe.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..2b40e8bc35055616f06de08151165f9675d5f690
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_24_layers_w_view_pe.yaml
@@ -0,0 +1,26 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_24_layers_ifr_w_view_pe"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
+  # Maximum number of views for positional encoding
+  max_num_views_for_pe: 1000
+  # Use random indices within range (1, max_num_views_for_pe) for positional encoding of non reference views
+  use_rand_idx_pe_for_non_reference_views: True
diff --git a/configs/model/info_sharing/aat_ifr_48_layers.yaml b/configs/model/info_sharing/aat_ifr_48_layers.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..20425dcf47583cb287f273278168f15e23e92f24
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_48_layers.yaml
@@ -0,0 +1,26 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_48_layers_ifr"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 23, 35]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "48_layers"
+  # Depth (this includes both frame-wise and gloabl attention layers)
+  depth: 48
+  # Feature dim (similar to ViT-Large)
+  dim: 1024
+  # Number of heads (similar to ViT-Large)
+  num_heads: 16
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/aat_ifr_48_layers_escaling.yaml b/configs/model/info_sharing/aat_ifr_48_layers_escaling.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..838837bc11c10823dfc39163ae4efb0ffe5ac2bb
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_48_layers_escaling.yaml
@@ -0,0 +1,28 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_48_layers_ifr"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 23, 35]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "48_layers"
+  # Depth (this includes both frame-wise and gloabl attention layers)
+  depth: 48
+  # Feature dim (similar to ViT-Large)
+  dim: 1024
+  # Number of heads (similar to ViT-Large)
+  num_heads: 16
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: True
+  # Scale Entropy in Attention
+  use_entropy_scaling: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/aat_ifr_48_layers_no_ref_view.yaml b/configs/model/info_sharing/aat_ifr_48_layers_no_ref_view.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b7dfea0c27591afe6d13f477b110a3ffdf7444d6
--- /dev/null
+++ b/configs/model/info_sharing/aat_ifr_48_layers_no_ref_view.yaml
@@ -0,0 +1,26 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "alternating_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "aat_48_layers_ifr_no_ref_view"
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 23, 35]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "48_layers"
+  # Depth (this includes both frame-wise and gloabl attention layers)
+  depth: 48
+  # Feature dim (similar to ViT-Large)
+  dim: 1024
+  # Number of heads (similar to ViT-Large)
+  num_heads: 16
+  # Distinguish Reference and Non-Reference Views
+  distinguish_ref_and_non_ref_views: False
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/cat_ifr_dust3r.yaml b/configs/model/info_sharing/cat_ifr_dust3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..34d4a175450864a2fc9dea44460e4c3aea13bbcf
--- /dev/null
+++ b/configs/model/info_sharing/cat_ifr_dust3r.yaml
@@ -0,0 +1,18 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "cross_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: "RoPE100"
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "base_cat_ifr_dust3r"
+  # Number of views
+  num_views: 2
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [5, 8]
+  # Normalize intermediate features
+  norm_intermediate: False
+  # Load CroCo cross-attention transformer for DUSt3R Init
+  pretrained_checkpoint_path: '${machine.root_uniception_pretrained_checkpoints_dir}/info_sharing/cross_attn_transformer/Two_View_Cross_Attention_Transformer_CroCo.pth'
diff --git a/configs/model/info_sharing/gat_ifr_24_layers.yaml b/configs/model/info_sharing/gat_ifr_24_layers.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..66055d0aadcd66290128fa90aed2545c3182698c
--- /dev/null
+++ b/configs/model/info_sharing/gat_ifr_24_layers.yaml
@@ -0,0 +1,24 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "global_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "gat_24_layers_ifr"
+  # Maximum number of views for positional encoding
+  max_num_views: 1000
+  # Use random indices within range (1, max_num_views) for positional encoding of non reference views
+  use_rand_idx_pe_for_non_reference_views: True
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/info_sharing/gat_ifr_24_layers_escaling.yaml b/configs/model/info_sharing/gat_ifr_24_layers_escaling.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1032a73835cc44adc9f6f86801abede704c74418
--- /dev/null
+++ b/configs/model/info_sharing/gat_ifr_24_layers_escaling.yaml
@@ -0,0 +1,26 @@
+# Model type (Options: ["cross_attention", "global_attention", "alternating_attention"])
+model_type: "global_attention"
+# Model class type (Options: ["no_intermediate_features", "intermediate_features"])
+model_return_type: "intermediate_features"
+# Custom positional encoding (Options: ["RoPEfreq"], Callable Function, null)
+custom_positional_encoding: null
+# Module arguments
+module_args:
+  # Name of the info sharing module
+  name: "gat_24_layers_ifr"
+  # Maximum number of views for positional encoding
+  max_num_views: 1000
+  # Use random indices within range (1, max_num_views) for positional encoding of non reference views
+  use_rand_idx_pe_for_non_reference_views: True
+  # Indices of the intermediate features to be shared (indices start from 0)
+  indices: [11, 17]
+  # Normalize intermediate features
+  norm_intermediate: True
+  # Size string
+  size: "24_layers"
+  # Depth
+  depth: 24
+  # Scale Entropy in Attention
+  use_entropy_scaling: True
+  # Flag to indicate whether to use gradient checkpointing
+  gradient_checkpointing: False
diff --git a/configs/model/mapanything.yaml b/configs/model/mapanything.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d0ffa3d1b9b6aea267d91567092da5ccd28bcf91
--- /dev/null
+++ b/configs/model/mapanything.yaml
@@ -0,0 +1,18 @@
+defaults:
+  - default
+  - encoder: dinov2_large
+  - info_sharing: aat_ifr_24_layers
+  - pred_head: dpt_pose_scale
+  - task: images_only
+
+# String for model factory
+model_str: "mapanything"
+# Model config
+model_config:
+  name: "mapanything"
+  encoder_config: ${model.encoder}
+  info_sharing_config: ${model.info_sharing}
+  pred_head_config: ${model.pred_head}
+  geometric_input_config: ${model.task}
+# Image Normalization Type
+data_norm_type: ${model.encoder.data_norm_type}
diff --git a/configs/model/mapanything_ablations.yaml b/configs/model/mapanything_ablations.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..97b5bc5065b179863672a96514e05ca6510363ac
--- /dev/null
+++ b/configs/model/mapanything_ablations.yaml
@@ -0,0 +1,18 @@
+defaults:
+  - default
+  - encoder: dinov2_large
+  - info_sharing: aat_ifr_24_layers
+  - pred_head: dpt_pose
+  - task: images_only
+
+# String for model factory
+model_str: "mapanything_ablations"
+# Model config
+model_config:
+  name: "mapanything_ablations"
+  encoder_config: ${model.encoder}
+  info_sharing_config: ${model.info_sharing}
+  pred_head_config: ${model.pred_head}
+  geometric_input_config: ${model.task}
+# Image Normalization Type
+data_norm_type: ${model.encoder.data_norm_type}
diff --git a/configs/model/mast3r.yaml b/configs/model/mast3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..142e2171363396a8f14a7dc6e2867fd51a13b999
--- /dev/null
+++ b/configs/model/mast3r.yaml
@@ -0,0 +1,15 @@
+# String for model factory
+model_str: "mast3r"
+# Model config
+model_config:
+  name: "mast3r"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/MASt3R_ViTLarge_BaseDecoder_512_catmlpdpt_metric.pth"
+  # Cache dir
+  cache_dir: "${root_pretrained_checkpoints_dir}/mast3r_cache"
+# Image Normalization Type
+data_norm_type: "dust3r"
+# MASt3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/metric_dust3r.yaml b/configs/model/metric_dust3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a0d4760172caab0475c345a1a2cd4821978bb9fc
--- /dev/null
+++ b/configs/model/metric_dust3r.yaml
@@ -0,0 +1,23 @@
+# String for model factory
+model_str: "dust3r"
+# Model config
+model_config:
+  name: "metric_dust3r"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/MASt3R_ViTLarge_BaseDecoder_512_catmlpdpt_metric.pth"
+  # Scene graph for BA
+  scene_graph: "complete"
+  # Pairwise inference batch size
+  inference_batch_size: 32
+  # Global optim schedule
+  global_optim_schedule: "cosine"
+  # Global optim lr
+  global_optim_lr: 0.01
+  # Number of iterations for global optimization
+  global_optim_niter: 300
+# Image Normalization Type
+data_norm_type: "dust3r"
+# DUSt3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/modular_dust3r_512_dpt.yaml b/configs/model/modular_dust3r_512_dpt.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..5e5107b2b146749e1682607abfebf40cdc115c6e
--- /dev/null
+++ b/configs/model/modular_dust3r_512_dpt.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+  - encoder: croco_512_data_norm_dust3r
+  - info_sharing: cat_ifr_dust3r
+  - pred_head: dpt
+
+# String for model factory
+model_str: "modular_dust3r"
+# Model config
+model_config:
+  name: "dust3r_512_dpt"
+  encoder_config: ${model.encoder}
+  info_sharing_config: ${model.info_sharing}
+  pred_head_config: ${model.pred_head}
+# Image Normalization Type
+data_norm_type: "dust3r"
diff --git a/configs/model/moge_1.yaml b/configs/model/moge_1.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..22dfc1ddc6ac22182be28a580dd3a98004efbd66
--- /dev/null
+++ b/configs/model/moge_1.yaml
@@ -0,0 +1,17 @@
+# String for model factory
+model_str: "moge"
+# Model config
+model_config:
+  name: "moge-1"
+  # MoGe pre-trained model checkpoint string
+  model_string: "Ruicheng/moge-vitl"
+  # Load custom checkpoint
+  load_custom_ckpt: false
+  # Custom checkpoint path
+  custom_ckpt_path: null
+# Image Normalization Type
+data_norm_type: "identity"
+# MoGe checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/moge_2.yaml b/configs/model/moge_2.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..41950288350393d4166a3066e6c6fb9e7fcd00cb
--- /dev/null
+++ b/configs/model/moge_2.yaml
@@ -0,0 +1,17 @@
+# String for model factory
+model_str: "moge"
+# Model config
+model_config:
+  name: "moge-2"
+  # MoGe pre-trained model checkpoint string
+  model_string: "Ruicheng/moge-2-vitl"
+  # Load custom checkpoint
+  load_custom_ckpt: false
+  # Custom checkpoint path
+  custom_ckpt_path: null
+# Image Normalization Type
+data_norm_type: "identity"
+# MoGe checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/must3r.yaml b/configs/model/must3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..edbb51708162a556c2ae2ac819fd084e6d1ec897
--- /dev/null
+++ b/configs/model/must3r.yaml
@@ -0,0 +1,15 @@
+# String for model factory
+model_str: "must3r"
+# Model config
+model_config:
+  name: "must3r"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/MUSt3R_512.pth"
+  # Retrieval Checkpoint path
+  retrieval_ckpt_path: "${root_pretrained_checkpoints_dir}/MUSt3R_512_retrieval_trainingfree.pth"
+# Image Normalization Type
+data_norm_type: "dust3r"
+# MASt3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/pi3.yaml b/configs/model/pi3.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..19afea09be091071f095825dddd3d2d5cc91a518
--- /dev/null
+++ b/configs/model/pi3.yaml
@@ -0,0 +1,13 @@
+# String for model factory
+model_str: "pi3"
+# Model config
+model_config:
+  name: "pi3"
+  # Load pre-trained weights
+  load_pretrained_weights: true
+# Image Normalization Type
+data_norm_type: "identity"
+# Pi3 checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: False
diff --git a/configs/model/pow3r.yaml b/configs/model/pow3r.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1c13ea3cead807f93ea7e6b809c4017d6b336293
--- /dev/null
+++ b/configs/model/pow3r.yaml
@@ -0,0 +1,19 @@
+defaults:
+  - default
+  - task: images_only
+
+# String for model factory
+model_str: "pow3r"
+# Model config
+model_config:
+  name: "pow3r"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/Pow3R_ViTLarge_BaseDecoder_512_linear.pth"
+  # Geometric input config
+  geometric_input_config: ${model.task}
+# Image Normalization Type
+data_norm_type: "dust3r"
+# Pow3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/pow3r_ba.yaml b/configs/model/pow3r_ba.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..8b232df603420878fa290998feff97c457cd59d5
--- /dev/null
+++ b/configs/model/pow3r_ba.yaml
@@ -0,0 +1,29 @@
+defaults:
+  - default
+  - task: images_only
+
+# String for model factory
+model_str: "pow3r_ba"
+# Model config
+model_config:
+  name: "pow3r_ba"
+  # Checkpoint path
+  ckpt_path: "${root_pretrained_checkpoints_dir}/Pow3R_ViTLarge_BaseDecoder_512_linear.pth"
+  # Geometric input config
+  geometric_input_config: ${model.task}
+  # Scene graph for BA
+  scene_graph: "complete"
+  # Pairwise inference batch size
+  inference_batch_size: 32
+  # Global optim schedule
+  global_optim_schedule: "cosine"
+  # Global optim lr
+  global_optim_lr: 0.01
+  # Number of iterations for global optimization
+  global_optim_niter: 300
+# Image Normalization Type
+data_norm_type: "dust3r"
+# Pow3R checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask.yaml b/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..61c1f1912bdeac50f23bd83b4678a7c3d551d28d
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask.yaml
@@ -0,0 +1,22 @@
+# Camera Frame Pointmap + Global Camera Pose (Trans + Quats) + Confidence + Mask
+input_dim: 5
+scene_rep_dim: 3
+type: "campointmap+pose+confidence+mask"
+scene_rep_type: "campointmap+pose"
+dense_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  pointmap_mode: "z_exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask_scale.yaml b/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..798b7f6dd2f43806dd0edceac168892977499052
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/campointmap_pose_confidence_mask_scale.yaml
@@ -0,0 +1,27 @@
+# Camera Frame Pointmap + Global Camera Pose (Trans + Quats) + Confidence + Mask + Scene-wide Metric Scaling Factor
+input_dim: 5
+scene_rep_dim: 3
+type: "campointmap+pose+confidence+mask"
+scene_rep_type: "campointmap+pose"
+dense_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  pointmap_mode: "z_exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
+scale_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  mode: "exp"
+  vmin: 1e-08
+  vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/pointmap_confidence.yaml b/configs/model/pred_head/adaptor_config/pointmap_confidence.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..50b89d8d2f9995143a8821198d2131015a413b9d
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/pointmap_confidence.yaml
@@ -0,0 +1,13 @@
+# Pointmap + Confidence
+input_dim: 4
+scene_rep_dim: 3
+type: "pointmap+confidence"
+scene_rep_type: "pointmap"
+init_dict:
+  name: "pointmap+confidence"
+  pointmap_mode: "exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/pointmap_confidence_mask.yaml b/configs/model/pred_head/adaptor_config/pointmap_confidence_mask.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..7d4acb17e456ab5883950458749c6f1b2773cb42
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/pointmap_confidence_mask.yaml
@@ -0,0 +1,13 @@
+# Pointmap + Confidence + Mask
+input_dim: 5
+scene_rep_dim: 3
+type: "pointmap+confidence+mask"
+scene_rep_type: "pointmap"
+init_dict:
+  name: "pointmap+confidence+mask"
+  pointmap_mode: "exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/pointmap_confidence_mask_scale.yaml b/configs/model/pred_head/adaptor_config/pointmap_confidence_mask_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3788cad2fd5d59e775f6c9a7143a3d7ea1ff7a86
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/pointmap_confidence_mask_scale.yaml
@@ -0,0 +1,18 @@
+# Pointmap + Confidence + Mask + Scene-wide Metric Scaling Factor
+input_dim: 5
+scene_rep_dim: 3
+type: "pointmap+confidence+mask"
+scene_rep_type: "pointmap"
+init_dict:
+  name: "pointmap+confidence+mask"
+  pointmap_mode: "exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+scale_pred_init_dict:
+  name: "campointmap+pose+confidence+mask+scale"
+  mode: "exp"
+  vmin: 1e-08
+  vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/pointmap_factored_raydirs_depth_pose_confidence_mask_scale.yaml b/configs/model/pred_head/adaptor_config/pointmap_factored_raydirs_depth_pose_confidence_mask_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c0597b92436a4dbef96399cc919f46369d76c596
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/pointmap_factored_raydirs_depth_pose_confidence_mask_scale.yaml
@@ -0,0 +1,39 @@
+# Global Pointmaps + Ray Directions on Unit Sphere + Depth along Ray + Global Camera Pose (Trans + Quats) + Confidence + Mask + Global Metric Scaling Factor
+input_dim: 9
+scene_rep_dim: 7
+type: "pointmap+raydirs+depth+pose+confidence+mask"
+scene_rep_type: "pointmap+raydirs+depth+pose"
+dense_pred_init_dict:
+  name: "pointmap+raydirs+depth+pose+confidence+mask+scale"
+  pointmap_mode: "exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  ray_directions_mode: "linear"
+  ray_directions_normalize_to_unit_sphere: true
+  ray_directions_normalize_to_unit_image_plane: false
+  ray_directions_vmin: ${special_float:"-inf"}
+  ray_directions_vmax: ${special_float:"inf"}
+  ray_directions_clamp_min_of_z_dir: false
+  ray_directions_z_dir_min: ${special_float:"-inf"}
+  depth_mode: "exp"
+  depth_vmin: 0
+  depth_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
+scale_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  mode: "exp"
+  vmin: 1e-08
+  vmax: ${special_float:"inf"}
+# Flag to decide what representaion to use for global pointmaps
+use_factored_predictions_for_global_pointmaps: true
diff --git a/configs/model/pred_head/adaptor_config/pointmap_raydirs_depth_pose_confidence_mask_scale.yaml b/configs/model/pred_head/adaptor_config/pointmap_raydirs_depth_pose_confidence_mask_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c64d7e630c4d6f78f63413cb0d60397eded249aa
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/pointmap_raydirs_depth_pose_confidence_mask_scale.yaml
@@ -0,0 +1,39 @@
+# Global Pointmaps + Ray Directions on Unit Sphere + Depth along Ray + Global Camera Pose (Trans + Quats) + Confidence + Mask + Global Metric Scaling Factor
+input_dim: 9
+scene_rep_dim: 7
+type: "pointmap+raydirs+depth+pose+confidence+mask"
+scene_rep_type: "pointmap+raydirs+depth+pose"
+dense_pred_init_dict:
+  name: "pointmap+raydirs+depth+pose+confidence+mask+scale"
+  pointmap_mode: "exp"
+  pointmap_vmin: ${special_float:"-inf"}
+  pointmap_vmax: ${special_float:"inf"}
+  ray_directions_mode: "linear"
+  ray_directions_normalize_to_unit_sphere: true
+  ray_directions_normalize_to_unit_image_plane: false
+  ray_directions_vmin: ${special_float:"-inf"}
+  ray_directions_vmax: ${special_float:"inf"}
+  ray_directions_clamp_min_of_z_dir: false
+  ray_directions_z_dir_min: ${special_float:"-inf"}
+  depth_mode: "exp"
+  depth_vmin: 0
+  depth_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
+scale_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  mode: "exp"
+  vmin: 1e-08
+  vmax: ${special_float:"inf"}
+# Flag to decide what representaion to use for global pointmaps
+use_factored_predictions_for_global_pointmaps: false
diff --git a/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask.yaml b/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..14cc198b36bec63341a121ab48e4db848f92f0cf
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask.yaml
@@ -0,0 +1,29 @@
+# Ray Directions on Unit Sphere + Depth along Ray + Global Camera Pose (Trans + Quats) + Confidence + Mask
+input_dim: 6
+scene_rep_dim: 4
+type: "raydirs+depth+pose+confidence+mask"
+scene_rep_type: "raydirs+depth+pose"
+dense_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask"
+  ray_directions_mode: "linear"
+  ray_directions_normalize_to_unit_sphere: true
+  ray_directions_normalize_to_unit_image_plane: false
+  ray_directions_vmin: ${special_float:"-inf"}
+  ray_directions_vmax: ${special_float:"inf"}
+  ray_directions_clamp_min_of_z_dir: false
+  ray_directions_z_dir_min: ${special_float:"-inf"}
+  depth_mode: "exp"
+  depth_vmin: 0
+  depth_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask_scale.yaml b/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3b96c477e3c3dc56f34a06deb4b3fd758a0f7d24
--- /dev/null
+++ b/configs/model/pred_head/adaptor_config/raydirs_depth_pose_confidence_mask_scale.yaml
@@ -0,0 +1,34 @@
+# Ray Directions on Unit Sphere + Depth along Ray + Global Camera Pose (Trans + Quats) + Confidence + Mask + Global Metric Scaling Factor
+input_dim: 6
+scene_rep_dim: 4
+type: "raydirs+depth+pose+confidence+mask"
+scene_rep_type: "raydirs+depth+pose"
+dense_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  ray_directions_mode: "linear"
+  ray_directions_normalize_to_unit_sphere: true
+  ray_directions_normalize_to_unit_image_plane: false
+  ray_directions_vmin: ${special_float:"-inf"}
+  ray_directions_vmax: ${special_float:"inf"}
+  ray_directions_clamp_min_of_z_dir: false
+  ray_directions_z_dir_min: ${special_float:"-inf"}
+  depth_mode: "exp"
+  depth_vmin: 0
+  depth_vmax: ${special_float:"inf"}
+  confidence_type: "exp"
+  confidence_vmin: 1
+  confidence_vmax: ${special_float:"inf"}
+pose_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  cam_trans_mode: "linear"
+  cam_trans_vmin: ${special_float:"-inf"}
+  cam_trans_vmax: ${special_float:"inf"}
+  quaternions_mode: "linear"
+  quaternions_normalize: true
+  quaternions_vmin: ${special_float:"-inf"}
+  quaternions_vmax: ${special_float:"inf"}
+scale_pred_init_dict:
+  name: "raydirs+depth+pose+confidence+mask+scale"
+  mode: "exp"
+  vmin: 1e-08
+  vmax: ${special_float:"inf"}
diff --git a/configs/model/pred_head/dpt.yaml b/configs/model/pred_head/dpt.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d15c69b9551dccba3837634b3c37ce73231c98b9
--- /dev/null
+++ b/configs/model/pred_head/dpt.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - adaptor_config: pointmap_confidence
+
+type: "dpt"
+feature_head:
+  feature_dim: 256
+  hooks: [0, 1, 2, 3]
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+regressor_head:
+  output_dim: ${model.pred_head.adaptor_config.input_dim}
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+adaptor_type: ${model.pred_head.adaptor_config.type}
+adaptor: ${model.pred_head.adaptor_config.init_dict}
+# Flag to indicate whether to use gradient checkpointing
+gradient_checkpointing: False
diff --git a/configs/model/pred_head/dpt_pose.yaml b/configs/model/pred_head/dpt_pose.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..e4ae286cd5c39e08081c8c0118bbeececcf9aaaa
--- /dev/null
+++ b/configs/model/pred_head/dpt_pose.yaml
@@ -0,0 +1,19 @@
+defaults:
+  - adaptor_config: raydirs_depth_pose_confidence_mask
+
+type: "dpt+pose"
+feature_head:
+  feature_dim: 256
+  hooks: [0, 1, 2, 3]
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+regressor_head:
+  output_dim: ${model.pred_head.adaptor_config.input_dim}
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+pose_head:
+  num_resconv_block: 2
+  rot_representation_dim: 4
+adaptor_type: ${model.pred_head.adaptor_config.type}
+dpt_adaptor: ${model.pred_head.adaptor_config.dense_pred_init_dict}
+pose_adaptor: ${model.pred_head.adaptor_config.pose_pred_init_dict}
+# Flag to indicate whether to use gradient checkpointing
+gradient_checkpointing: False
diff --git a/configs/model/pred_head/dpt_pose_scale.yaml b/configs/model/pred_head/dpt_pose_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..223f7291702cf5380d8e86f3a79f3fce90314534
--- /dev/null
+++ b/configs/model/pred_head/dpt_pose_scale.yaml
@@ -0,0 +1,22 @@
+defaults:
+  - adaptor_config: raydirs_depth_pose_confidence_mask_scale
+
+type: "dpt+pose"
+feature_head:
+  feature_dim: 256
+  hooks: [0, 1, 2, 3]
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+regressor_head:
+  output_dim: ${model.pred_head.adaptor_config.input_dim}
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+pose_head:
+  num_resconv_block: 2
+  rot_representation_dim: 4
+scale_head:
+  output_dim: 1
+adaptor_type: ${model.pred_head.adaptor_config.type}
+dpt_adaptor: ${model.pred_head.adaptor_config.dense_pred_init_dict}
+pose_adaptor: ${model.pred_head.adaptor_config.pose_pred_init_dict}
+scale_adaptor: ${model.pred_head.adaptor_config.scale_pred_init_dict}
+# Flag to indicate whether to use gradient checkpointing
+gradient_checkpointing: False
diff --git a/configs/model/pred_head/dpt_scale.yaml b/configs/model/pred_head/dpt_scale.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..0e3f3b66ef17d0c39eab7a0a670e9b9696e49cce
--- /dev/null
+++ b/configs/model/pred_head/dpt_scale.yaml
@@ -0,0 +1,18 @@
+defaults:
+  - adaptor_config: pointmap_confidence_mask_scale
+
+type: "dpt"
+feature_head:
+  feature_dim: 256
+  hooks: [0, 1, 2, 3]
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+regressor_head:
+  output_dim: ${model.pred_head.adaptor_config.input_dim}
+  checkpoint_gradient: ${model.pred_head.gradient_checkpointing}
+scale_head:
+  output_dim: 1
+adaptor_type: ${model.pred_head.adaptor_config.type}
+adaptor: ${model.pred_head.adaptor_config.init_dict}
+scale_adaptor: ${model.pred_head.adaptor_config.scale_pred_init_dict}
+# Flag to indicate whether to use gradient checkpointing
+gradient_checkpointing: False
diff --git a/configs/model/task/aug_training.yaml b/configs/model/task/aug_training.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..ab5bcda9ca27f48d0bde5eabd7bc65ec9942023b
--- /dev/null
+++ b/configs/model/task/aug_training.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 0.9
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0.05
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 0.5
+# Probability of Geometric Inputs with Depths
+depth_prob: 0.5
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0.5
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0.5
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale quantities for the input metric high quality gt depth
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0.05
+# Probability for skipping input of the metric scale quantities for the input metric pose
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0.05
diff --git a/configs/model/task/calibrated_sfm.yaml b/configs/model/task/calibrated_sfm.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..36a032bd80f6f85d382e7228e7b41fbbcc215210
--- /dev/null
+++ b/configs/model/task/calibrated_sfm.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/default.yaml b/configs/model/task/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..59b8f6906605b537a52a920f0410f1bb3f24679d
--- /dev/null
+++ b/configs/model/task/default.yaml
@@ -0,0 +1,27 @@
+# Ray Directions Encoder Config
+ray_dirs_encoder_config:
+  name: "ray_dirs_encoder"
+  in_chans: 3
+  encoder_str: "dense_rep_encoder"
+  apply_pe: false
+# Depth Encoder Config
+depth_encoder_config:
+  name: "depth_encoder"
+  in_chans: 1
+  encoder_str: "dense_rep_encoder"
+  apply_pe: false
+# Cam Rotation (Quats) Encoder Config
+cam_rot_encoder_config:
+  name: "cam_rot_quats_encoder"
+  in_chans: 4
+  encoder_str: "global_rep_encoder"
+# Cam Translation Encoder Config
+cam_trans_encoder_config:
+  name: "cam_trans_encoder"
+  in_chans: 3
+  encoder_str: "global_rep_encoder"
+# Scale Encoder Config
+scale_encoder_config:
+  name: "scale_encoder"
+  in_chans: 1
+  encoder_str: "global_rep_encoder"
diff --git a/configs/model/task/depth_completion.yaml b/configs/model/task/depth_completion.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4a1f69ac3c9f0e9e3dd5d85643fdb9384231184c
--- /dev/null
+++ b/configs/model/task/depth_completion.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 1
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/images_only.yaml b/configs/model/task/images_only.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a964709223cbc31147f9424c6b4d0ae790c5aeaa
--- /dev/null
+++ b/configs/model/task/images_only.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 0
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 1
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 0
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/mvs.yaml b/configs/model/task/mvs.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..55a9c6c1111ddc5f29b8f2d602ba6262de93a1c7
--- /dev/null
+++ b/configs/model/task/mvs.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/mvs_non_metric.yaml b/configs/model/task/mvs_non_metric.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3e82b6170d9f72d0aaa0f3088fb215176ece9919
--- /dev/null
+++ b/configs/model/task/mvs_non_metric.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/mvs_training.yaml b/configs/model/task/mvs_training.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d2d6b151eba1a4b713087a07e5a18ff1cda4aee9
--- /dev/null
+++ b/configs/model/task/mvs_training.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale quantities for the input metric high quality gt depth
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale quantities for the input metric pose
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0.05
diff --git a/configs/model/task/non_metric_poses_metric_depth.yaml b/configs/model/task/non_metric_poses_metric_depth.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..118a92a0109d96aff58f54d5a9d5a4ce9ae4f820
--- /dev/null
+++ b/configs/model/task/non_metric_poses_metric_depth.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/non_metric_poses_metric_depth_sparse.yaml b/configs/model/task/non_metric_poses_metric_depth_sparse.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..8dd18678da962f48547359aebc1de46192cefdb7
--- /dev/null
+++ b/configs/model/task/non_metric_poses_metric_depth_sparse.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 1
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/non_metric_poses_non_metric_depth.yaml b/configs/model/task/non_metric_poses_non_metric_depth.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b40c966ed86274d75d6386bfb73a7f50459ce80e
--- /dev/null
+++ b/configs/model/task/non_metric_poses_non_metric_depth.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 1
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/non_metric_poses_non_metric_depth_sparse.yaml b/configs/model/task/non_metric_poses_non_metric_depth_sparse.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..07368b82120063ff57e567da7f4f2d51ba7facae
--- /dev/null
+++ b/configs/model/task/non_metric_poses_non_metric_depth_sparse.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 1
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 1
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/pass_through.yaml b/configs/model/task/pass_through.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..cd117090ef8144a218390d3e53adbffaa481c0eb
--- /dev/null
+++ b/configs/model/task/pass_through.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/posed_sfm.yaml b/configs/model/task/posed_sfm.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..2de61efcd26b6210b297c4197fcff625021c7087
--- /dev/null
+++ b/configs/model/task/posed_sfm.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 0
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/posed_sfm_non_metric.yaml b/configs/model/task/posed_sfm_non_metric.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..05e96622612d5f16d95cd603d12f4ce3f12ce373
--- /dev/null
+++ b/configs/model/task/posed_sfm_non_metric.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 0
+# Probability of Geometric Inputs with Depths
+depth_prob: 0
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 1
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 1
diff --git a/configs/model/task/registration.yaml b/configs/model/task/registration.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a34588baa5910781a1af715394886390e916b790
--- /dev/null
+++ b/configs/model/task/registration.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/registration_sparse.yaml b/configs/model/task/registration_sparse.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..c2fd0da53475885bd2df1b69071c99a2360c96da
--- /dev/null
+++ b/configs/model/task/registration_sparse.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 1
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale norm factor for the input metric high quality gt depth
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0
+# Probability for skipping input of the metric scale norm factor for the input metric pose
+# If 0, the metric scale norm factor will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/task/registration_training.yaml b/configs/model/task/registration_training.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..f30435d26f93bc1dc4f36ac87c42e44f6e71f39e
--- /dev/null
+++ b/configs/model/task/registration_training.yaml
@@ -0,0 +1,23 @@
+defaults:
+  - default
+
+# Overall Probability of Geometric Inputs
+overall_prob: 1
+# Dropout Probability of Geometric Inputs (for each sample across batch size and number of views)
+dropout_prob: 0
+# Probability of Geometric Inputs with Ray Directions
+ray_dirs_prob: 1
+# Probability of Geometric Inputs with Depths
+depth_prob: 1
+# Probability of Geometric Inputs with Camera Poses
+cam_prob: 0
+# Probability of sparsely sampling the high quality gt depth
+sparse_depth_prob: 0.5
+# Percentage of the valid depth to remove if the probability of using sparse depth is greater than 0 (Range: [0, 1])
+sparsification_removal_percent: 0.9
+# Probability for skipping input of the metric scale quantities for the input metric high quality gt depth
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+depth_scale_norm_all_prob: 0.05
+# Probability for skipping input of the metric scale quantities for the input metric pose
+# If 0, the metric scale quantities will be provided as input to the model for all the metric scale conditionings
+pose_scale_norm_all_prob: 0
diff --git a/configs/model/vggt.yaml b/configs/model/vggt.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..fe637acc2f7d37b19480e8e63fe197b2567fe3f0
--- /dev/null
+++ b/configs/model/vggt.yaml
@@ -0,0 +1,17 @@
+# String for model factory
+model_str: "vggt"
+# Model config
+model_config:
+  name: "vggt"
+  # Load pre-trained weights
+  load_pretrained_weights: true
+  # Load custom checkpoint
+  load_custom_ckpt: false
+  # Custom checkpoint path
+  custom_ckpt_path: null
+# Image Normalization Type
+data_norm_type: "identity"
+# VGGT checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/vggt_commercial.yaml b/configs/model/vggt_commercial.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3947977319fb5c16d275c71434f76f82fd7f69be
--- /dev/null
+++ b/configs/model/vggt_commercial.yaml
@@ -0,0 +1,17 @@
+# String for model factory
+model_str: "vggt"
+# Model config
+model_config:
+  name: "vggt"
+  # Load pre-trained weights
+  load_pretrained_weights: true
+  # Load custom checkpoint
+  load_custom_ckpt: true
+  # Custom checkpoint path
+  custom_ckpt_path: "${root_pretrained_checkpoints_dir}/vggt_1B_commercial.pt"
+# Image Normalization Type
+data_norm_type: "identity"
+# VGGT checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/model/vggt_non_pretrained.yaml b/configs/model/vggt_non_pretrained.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..7c17f88844eed48f5c009af417ecaed5b243c624
--- /dev/null
+++ b/configs/model/vggt_non_pretrained.yaml
@@ -0,0 +1,17 @@
+# String for model factory
+model_str: "vggt"
+# Model config
+model_config:
+  name: "vggt"
+  # Load pre-trained weights
+  load_pretrained_weights: false
+  # Load custom checkpoint
+  load_custom_ckpt: false
+  # Custom checkpoint path
+  custom_ckpt_path: null
+# Image Normalization Type
+data_norm_type: "identity"
+# VGGT checkpoint is already loaded in the inference wrapper
+pretrained: null
+# Torch hub force reload
+torch_hub_force_reload: false
diff --git a/configs/rmvd_benchmark.yaml b/configs/rmvd_benchmark.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3c07521c8b47846bb98ae8a40e340b416c445db3
--- /dev/null
+++ b/configs/rmvd_benchmark.yaml
@@ -0,0 +1,34 @@
+defaults:
+  - machine: aws
+  - model: default
+  - dataset: default
+  - _self_
+
+# Path Settings
+output_dir: ${hydra:run.dir}
+external_benchmark_data_root_data_dir: ${machine.external_benchmark_data_root_data_dir}
+mapanything_dataset_metadata_dir: ${machine.mapanything_dataset_metadata_dir}
+root_pretrained_checkpoints_dir: ${machine.root_pretrained_checkpoints_dir}
+root_experiments_dir: ${machine.root_experiments_dir}
+root_uniception_pretrained_checkpoints_dir: ${machine.root_uniception_pretrained_checkpoints_dir}
+
+### Benchmarking args
+seed: 0
+# Disable CUDNN Benchmark (Disable for variable resolution & number of view training)
+disable_cudnn_benchmark: true
+# Batch size for inference (Metrics are computed per multi-view set and averaged, not per batch of multi-view sets)
+batch_size: 10
+# Use mixed precision for inference
+amp: 1
+# Floating point type to use for mixed precision
+amp_dtype: "bf16"
+# Choose from eth3d, kitti, scannet
+eval_dataset: eth3d
+# Choose from img, img+intrinsics, img+intrinsics+pose
+evaluation_conditioning: img
+# Choose from "median", "none"
+evaluation_alignment: median
+# Choose from "multi_view", "single_view"
+evaluation_views: multi_view
+# Resolution to inference the selected model.
+evaluation_resolution: ${dataset.resolution_options.518_1_33_ar}
diff --git a/configs/train.yaml b/configs/train.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..b35f6b9e9d848422d90d92db4807b3cf102c1961
--- /dev/null
+++ b/configs/train.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - machine: aws
+  - model: default
+  - dataset: default
+  - loss: default
+  - train_params: default
+  - distributed: default
+  - _self_
+
+output_dir: ${hydra:run.dir}
+root_data_dir: ${machine.root_data_dir}
+mapanything_dataset_metadata_dir: ${machine.mapanything_dataset_metadata_dir}
+root_pretrained_checkpoints_dir: ${machine.root_pretrained_checkpoints_dir}
+root_experiments_dir: ${machine.root_experiments_dir}
+root_uniception_pretrained_checkpoints_dir: ${machine.root_uniception_pretrained_checkpoints_dir}
diff --git a/configs/train_params/default.yaml b/configs/train_params/default.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..28af2bbf8810776b9ea02f14fb6f9264c9ee986a
--- /dev/null
+++ b/configs/train_params/default.yaml
@@ -0,0 +1,41 @@
+# Random Seed
+seed: 0
+# Maximum number of images per GPU (changes based on available GPU memory)
+max_num_of_imgs_per_gpu: 48
+# Accumulate gradient iterations (for increasing the effective batch size under memory constraints)
+accum_iter: 1
+# Maximum number of epochs for the scheduler
+epochs: 100
+## Default Optimizer parameters
+# Learning rate (absolute lr)
+lr: 0.0001
+# Lower lr bound for cyclic schedulers that hit 0
+min_lr: 1e-06
+# Epochs to warmup LR
+warmup_epochs: 10
+# Weight decay
+weight_decay: 0.05
+# LR schedule type
+schedule_type: "linear_warmup_half_cycle_cosine_decay"
+# Warn if model params are not in the below submodule_configs
+warn_not_in_submodule: False
+# Optimizer parameters specific to submodules
+submodule_configs: {}
+# Use Automatic Mixed Precision for pretraining
+amp: 1
+# Floating point type to use for mixed precision training
+amp_dtype: "bf16"
+# Disable CUDNN Benchmark (Disable for variable resolution & number of view training)
+disable_cudnn_benchmark: true
+# Freeze the validation samples across all epochs
+freeze_val_samples_across_all_epochs: true
+# Test loss evaluation frequency
+eval_freq: 1
+# Frequency (number of epochs) to save checkpoint in checkpoint-last.pth
+save_freq: 1
+# Frequency (number of epochs) to save checkpoint in checkpoint-%d.pth
+keep_freq: 10
+# Frequence (number of iterations) to print infos while training (includes tensorboard logging)
+print_freq: 20
+# Resume Training from last checkpoint
+resume: True
diff --git a/configs/train_params/finetune_with_lower_encoder_lr.yaml b/configs/train_params/finetune_with_lower_encoder_lr.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..54710c25f7ec51b9cc8063b6b82dc9ae08e1b913
--- /dev/null
+++ b/configs/train_params/finetune_with_lower_encoder_lr.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+
+# Use 20x lower lr for finetuning
+lr: 5e-06
+min_lr: 5e-08
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # Encoder
+  encoder:
+    lr: 2.5e-07
+    min_lr: 2.5e-09
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}
diff --git a/configs/train_params/finetune_with_lower_encoder_lr_64g.yaml b/configs/train_params/finetune_with_lower_encoder_lr_64g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..8dc9cf2ceeea3dcc609bcaae5a495ba337c8fe4e
--- /dev/null
+++ b/configs/train_params/finetune_with_lower_encoder_lr_64g.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+
+# Use 20x lower lr for finetuning
+lr: 1e-05
+min_lr: 1e-07
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # Encoder
+  encoder:
+    lr: 5e-07
+    min_lr: 5e-09
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}
diff --git a/configs/train_params/freeze_encoder.yaml b/configs/train_params/freeze_encoder.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..a20d13d08e82ae18b6f34403c614e121109fe442
--- /dev/null
+++ b/configs/train_params/freeze_encoder.yaml
@@ -0,0 +1,8 @@
+defaults:
+  - default
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # Encoder
+  encoder:
+    lr: 0
diff --git a/configs/train_params/lower_encoder_lr.yaml b/configs/train_params/lower_encoder_lr.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..4225312d1da1f4a9108e5cb97ea4a02b84b1c32e
--- /dev/null
+++ b/configs/train_params/lower_encoder_lr.yaml
@@ -0,0 +1,12 @@
+defaults:
+  - default
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # Encoder
+  encoder:
+    lr: 5e-06
+    min_lr: 5e-08
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}
diff --git a/configs/train_params/lower_encoder_lr_64g.yaml b/configs/train_params/lower_encoder_lr_64g.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..d7e426d217f67887cc0fc3d5d8bb1ef6312c1e4b
--- /dev/null
+++ b/configs/train_params/lower_encoder_lr_64g.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+
+# Use 2x higher lr for 8x higher effective batch size
+lr: 2e-04
+min_lr: 2e-07
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # Encoder
+  encoder:
+    lr: 1e-05
+    min_lr: 1e-08
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}
diff --git a/configs/train_params/moge2_finetune.yaml b/configs/train_params/moge2_finetune.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..f8cedc03f8bd95a1ddcef4823d3a3fc3bf91b1ab
--- /dev/null
+++ b/configs/train_params/moge2_finetune.yaml
@@ -0,0 +1,6 @@
+defaults:
+  - default
+
+# Use lower lr for finetuning
+lr: 1e-05
+min_lr: 1e-07
diff --git a/configs/train_params/pi3_finetune.yaml b/configs/train_params/pi3_finetune.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..cd03510fc3b1ded194965d5230a6763ed440ef82
--- /dev/null
+++ b/configs/train_params/pi3_finetune.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+
+# Use lower lr for finetuning
+lr: 1e-05
+min_lr: 1e-07
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # DINOv2
+  model.encoder:
+    lr: 5e-07
+    min_lr: 5e-09
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}
diff --git a/configs/train_params/vggt_finetune.yaml b/configs/train_params/vggt_finetune.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..1c0ae4ee46d62cbf0fb07188d4bfbe64c70e34a4
--- /dev/null
+++ b/configs/train_params/vggt_finetune.yaml
@@ -0,0 +1,16 @@
+defaults:
+  - default
+
+# Use lower lr for finetuning
+lr: 1e-05
+min_lr: 1e-07
+
+# Optimizer parameters specific to submodules
+submodule_configs:
+  # DINOv2
+  model.aggregator.patch_embed:
+    lr: 5e-07
+    min_lr: 5e-09
+    warmup_epochs: ${train_params.warmup_epochs}
+    weight_decay: ${train_params.weight_decay}
+    schedule_type: ${train_params.schedule_type}