joonhyun23452 commited on
Commit
43afdb7
·
1 Parent(s): 7f3a3f1

fix: add gdown for model checkpoints

Browse files
Files changed (3) hide show
  1. app.py +9 -3
  2. models/__init__.py +0 -0
  3. requirements.txt +1 -0
app.py CHANGED
@@ -4,6 +4,7 @@ import os
4
  import wget
5
  import gradio as gr
6
  import numpy as np
 
7
  from argparse import Namespace
8
  try:
9
  import detectron2
@@ -33,6 +34,11 @@ lvis_train_cat_info_path = "datasets/metadata/lvis_v1_train_cat_info.json"
33
  if not os.path.exists(lvis_train_cat_info_path):
34
  wget.download("https://docs.google.com/uc?export=download&id=17WmkAJYBK4xT-YkiXLcwIWmtfulSUtmO", out=lvis_train_cat_info_path)
35
 
 
 
 
 
 
36
  args = Namespace(
37
  base_cat_threshold=0.9,
38
  confidence_threshold=0.0,
@@ -40,13 +46,13 @@ args = Namespace(
40
  cpu=not torch.cuda.is_available(),
41
  custom_vocabulary='headphone,webcam,paper,coffe',
42
  input=['.assets/desk.jpg'],
43
- opts=['MODEL.WEIGHTS', 'models/proxydet_swinb_w_inl.pth'],
44
  output='out.jpg',
45
  pred_all_class=False,
46
  video_input=None,
47
  vocabulary='custom',
48
  webcam=None,
49
- zeroshot_weight_path='datasets/metadata/lvis_v1_clip_a+cname.npy'
50
  )
51
  cfg = setup_cfg(args)
52
  ovd_demo = VisualizationDemo(cfg, args)
@@ -90,7 +96,7 @@ How to use?
90
  - Simply upload an image and enter comma separated objects (e.g., "dog,cat,headphone") which you want to detect within the image.
91
  Parameters:
92
  - You can also use the score threshold slider to set a threshold to filter out low probability predictions.
93
- - adjust alpha and beta value for base and novel classes, respectively. These determine <b>how much importance will you assign to the scores sourced from our proposed detection head which is trained with our proxy-novel classes</b>
94
  """
95
  demo = gr.Interface(
96
  query_image,
 
4
  import wget
5
  import gradio as gr
6
  import numpy as np
7
+ import gdown
8
  from argparse import Namespace
9
  try:
10
  import detectron2
 
34
  if not os.path.exists(lvis_train_cat_info_path):
35
  wget.download("https://docs.google.com/uc?export=download&id=17WmkAJYBK4xT-YkiXLcwIWmtfulSUtmO", out=lvis_train_cat_info_path)
36
 
37
+ # download model
38
+ model_path = "models/proxydet_swinb_w_inl.pth"
39
+ if not os.path.exists(model_path):
40
+ gdown.download("https://docs.google.com/uc?export=download&id=17kUPoi-pEK7BlTBheGzWxe_DXJlg28qF", model_path)
41
+
42
  args = Namespace(
43
  base_cat_threshold=0.9,
44
  confidence_threshold=0.0,
 
46
  cpu=not torch.cuda.is_available(),
47
  custom_vocabulary='headphone,webcam,paper,coffe',
48
  input=['.assets/desk.jpg'],
49
+ opts=['MODEL.WEIGHTS', model_path],
50
  output='out.jpg',
51
  pred_all_class=False,
52
  video_input=None,
53
  vocabulary='custom',
54
  webcam=None,
55
+ zeroshot_weight_path=zs_weight_path
56
  )
57
  cfg = setup_cfg(args)
58
  ovd_demo = VisualizationDemo(cfg, args)
 
96
  - Simply upload an image and enter comma separated objects (e.g., "dog,cat,headphone") which you want to detect within the image.
97
  Parameters:
98
  - You can also use the score threshold slider to set a threshold to filter out low probability predictions.
99
+ - adjust alpha and beta value for base and novel classes, respectively. These determine <b>how much importance will you assign to the scores sourced from our proposed detection head which is trained with our proxy-novel classes</b>.
100
  """
101
  demo = gr.Interface(
102
  query_image,
models/__init__.py ADDED
File without changes
requirements.txt CHANGED
@@ -10,3 +10,4 @@ lvis
10
  wget==3.2
11
  nltk<=3.7
12
  numpy>=1.18.5
 
 
10
  wget==3.2
11
  nltk<=3.7
12
  numpy>=1.18.5
13
+ gdown==4.7.3