Datasets:

Modalities:
Video
Languages:
English
Tags:
video
License:
qingy2024 commited on
Commit
018892e
·
verified ·
1 Parent(s): dadfe61

Upload k600/train/train/generate_json_dataset.py with huggingface_hub

Browse files
k600/train/train/generate_json_dataset.py CHANGED
@@ -6,36 +6,29 @@ import random
6
  def generate_dataset_json(root_dir='.', output_file='dataset.json'):
7
  dataset = []
8
 
9
- # Get all directories in root_dir (each represents an action)
10
  actions = [d for d in os.listdir(root_dir) if os.path.isdir(os.path.join(root_dir, d))]
11
 
12
  for action in actions:
13
  action_dir = os.path.join(root_dir, action)
14
 
15
- # Find all mp4 files in this action directory
16
  video_files = glob(os.path.join(action_dir, '*.mp4'))
17
 
18
  for video_path in video_files:
19
- # Create relative path from root_dir
20
  rel_path = os.path.relpath(video_path, root_dir)
21
 
22
- # Create caption by capitalizing the action name
23
  caption = action.capitalize()
24
 
25
  dataset.append({
26
- "video": rel_path.replace('\\', '/'), # Use forward slashes for consistency
27
  "caption": caption
28
  })
29
 
30
- # Shuffle the dataset with seed 42 for reproducibility
31
  random.seed(42)
32
  random.shuffle(dataset)
33
 
34
- # Write to JSON file
35
  with open(output_file, 'w') as f:
36
  json.dump(dataset, f, indent=2)
37
 
38
  print(f"Dataset JSON file created at {output_file} with {len(dataset)} entries.")
39
 
40
- # Example usage:
41
- generate_dataset_json(root_dir='./', output_file='kinetics.json')
 
6
  def generate_dataset_json(root_dir='.', output_file='dataset.json'):
7
  dataset = []
8
 
 
9
  actions = [d for d in os.listdir(root_dir) if os.path.isdir(os.path.join(root_dir, d))]
10
 
11
  for action in actions:
12
  action_dir = os.path.join(root_dir, action)
13
 
 
14
  video_files = glob(os.path.join(action_dir, '*.mp4'))
15
 
16
  for video_path in video_files:
 
17
  rel_path = os.path.relpath(video_path, root_dir)
18
 
 
19
  caption = action.capitalize()
20
 
21
  dataset.append({
22
+ "video": rel_path.replace('\\', '/'),
23
  "caption": caption
24
  })
25
 
 
26
  random.seed(42)
27
  random.shuffle(dataset)
28
 
 
29
  with open(output_file, 'w') as f:
30
  json.dump(dataset, f, indent=2)
31
 
32
  print(f"Dataset JSON file created at {output_file} with {len(dataset)} entries.")
33
 
34
+ generate_dataset_json(root_dir='./', output_file='kinetics.json')