huggingface space

app.py

@@ -11,7 +11,7 @@ from webui.merge_config_gradio import merge_config_then_run
 import huggingface_hub
 import shutil
 import os
-
+import torch
 
 HF_TOKEN = os.getenv('HF_TOKEN')
 pipe = merge_config_then_run()
@@ -39,21 +39,10 @@ If you have any questions, please feel free to reach me out at <b>knightyxp@gmai
 """
 
 
+def update_layout_visibility(selected_num):
+    num = int(selected_num)
+    return [gr.update(visible=(i < num)) for i in range(len(layout_files))]
 
-def update_layout_visibility(num):
-    """
-    Given the user's selection (string) in ["2","3","4","5"],
-    return visibility updates for each of the 5 layout video inputs.
-    """
-    n = int(num)
-    # Show layout_file1 if n >= 1, layout_file2 if n >= 2, etc.
-    return [
-        gr.update(visible=(n >= 1)),
-        gr.update(visible=(n >= 2)),
-        gr.update(visible=(n >= 3)),
-        gr.update(visible=(n >= 4)),
-        gr.update(visible=(n >= 5))
-    ]
 
 with gr.Blocks(css='style.css') as demo:
     # gr.Markdown(TITLE)
@@ -138,55 +127,28 @@ with gr.Blocks(css='style.css') as demo:
                     info="Please select the number of editing areas"
                 )
 
-                # Put all layout-video components in one Row to display them horizontally.
+                # 使用循环生成所有的布局视频组件，并存到列表 layout_files 中
+                layout_files = []
                 with gr.Row():
-                    layout_file1 = gr.Video(
-                        label="Layout Video 1",
-                        type="numpy",
-                        format="mp4",
-                        visible=True
-                    )
-                    layout_file2 = gr.Video(
-                        label="Layout Video 2",
-                        type="numpy",
-                        format="mp4",
-                        visible=True
-                    )
-                    layout_file3 = gr.Video(
-                        label="Layout Video 3",
-                        type="numpy",
-                        format="mp4",
-                        visible=False
-                    )
-                    layout_file4 = gr.Video(
-                        label="Layout Video 4",
-                        type="numpy",
-                        format="mp4",
-                        visible=False
-                    )
-                    layout_file5 = gr.Video(
-                        label="Layout Video 5",
-                        type="numpy",
-                        format="mp4",
-                        visible=False
-                    )
+                    for i in range(5):
+                        video = gr.Video(
+                            label=f"Layout Video {i+1}",
+                            type="numpy",
+                            format="mp4",
+                            visible=(i < 2)  # 默认显示前两个
+                        )
+                        layout_files.append(video)
 
-                # Toggle visibility of the layout videos based on user selection
+                # 当 num_layouts 改变时，通过回调函数更新 layout_files 列表中各视频组件的 visible 属性
                 num_layouts.change(
                     fn=update_layout_visibility,
                     inputs=num_layouts,
-                    outputs=[
-                        layout_file1,
-                        layout_file2,
-                        layout_file3,
-                        layout_file4,
-                        layout_file5
-                    ]
+                    outputs=layout_files
                 )
 
                 prompt = gr.Textbox(label='Prompt',
                                     info='Change the prompt, and extract each local prompt in the editing prompts.\
-                                    the local prompt order should be same as layout masks order.)',
+                                    (the local prompt order should be same as layout masks order.)',
                                 )
 
                 model_id = gr.Dropdown(
@@ -198,11 +160,25 @@ with gr.Blocks(css='style.css') as demo:
                     value='stable-diffusion-v1-5/stable-diffusion-v1-5')
 
 
-            run_button = gr.Button('Generate')
-
         with gr.Column():
             result = gr.Video(label='Result')
             # result.style(height=512, width=512)
+            with gr.Accordion('Temporal Crop offset and Sampling Stride', open=False):
+                n_sample_frame = gr.Slider(label='Number of Frames',
+                                    minimum=0,
+                                    maximum=32,
+                                    step=1,
+                                    value=16)
+                sampling_rate = gr.Slider(label='sampling_rate',
+                                        minimum=0,
+                                        maximum=20,
+                                        step=1,
+                                        value=1)
+                start_sample_frame = gr.Number(label='Start frame in the video',
+                            value=0,
+                            precision=0)
+
+
             with gr.Row():  
                 control_list = ['dwpose', 'depth_zoe', 'depth_midas']
                 control_type = gr.Dropdown(
@@ -252,7 +228,9 @@ with gr.Blocks(css='style.css') as demo:
                     value=["1"],
                     info="Select one or more flatten resolution factors. Mapping: 1 -> 64, 2 -> 32 (64/2), 4 -> 16 (64/4), 8 -> 8 (64/8)."
                 )
-                
+
+
+            run_button = gr.Button('Generate')                
 
     with gr.Row():
         from example import style_example
@@ -278,25 +256,22 @@ with gr.Blocks(css='style.css') as demo:
         #             # cache_examples=os.getenv('SYSTEM') == 'spaces'
         #             )
     gr.Markdown(ARTICLE)
-    inputs = [
-              model_id,
-              user_input_video,
-              num_layouts,
-              layout_file1,
-              layout_file2,
-              layout_file3,
-              layout_file4,
-              layout_file5,
-              prompt,
-              model_id,
-              control_type,
-              dwpose_options,
-              controlnet_conditioning_scale,
-              use_pnp,
-              pnp_inject_steps,
-              flatten_res,
+    inputs = [user_input_video, num_layouts,
+        *layout_files,
+        prompt,
+        model_id,
+        n_sample_frame,
+        start_sample_frame,
+        sampling_rate,
+        control_type,
+        dwpose_options,
+        controlnet_conditioning_scale,
+        use_pnp,
+        pnp_inject_steps,
+        flatten_res,
     ]
     prompt.submit(fn=pipe.run, inputs=inputs, outputs=result)
     run_button.click(fn=pipe.run, inputs=inputs, outputs=result)
-
+    if device == 'cuda':
+        torch.cuda.empty_cache()
 demo.queue().launch()

config/demo_config.yaml

@@ -2,13 +2,13 @@ pretrained_model_path: "/home/xianyang/Data/code/FateZero/ckpt/stable-diffusion-
 logdir: ./result/run_two_man/instance_level/3cls_spider_polar_vis_cross_attn
 
 dataset_config:
-    path: "data/run_two_man/run_two_man_fr2"
-    prompt: 'Man in red hoddie and man in gray shirt are jogging in forest'
+    path: ""
+    prompt: ""
     n_sample_frame: 16
     start_sample_frame: 0
-    sampling_rate: 2
-    layout_mask_dir: "./data/run_two_man/layout_masks_fr2"
-    layout_mask_order: ['left_man_plus','right_man_plus','trees','trunk']
+    sampling_rate: 1
+    layout_mask_dir: ""
+    layout_mask_order: []
     negative_promot: "ugly, blurry, low res, unrealistic, unaesthetic"
 
 control_config:
@@ -34,7 +34,7 @@ editing_config:
     sample_seeds: [0]
     num_inference_steps: 50
     blending_percentage: 0
-    vis_cross_attn: True
+    vis_cross_attn: False
     #cluster_inversion_feature: True