This Python program allows you to manually select clips from a dataset, enabling you to discard those clips that don’t meet the expectations for your model.

To execute the program, run the following command:

python main.py --parent_dir {path}

Upon running, the tool will load the paths of all the clips in the specified directory. The paths are stored in a videos_path.txt file, allowing you to easily resume your work.

• To accept a video, simply press the Enter key.
• To decline a video, press 'r' or Esc key.
• To undo last action, press 'u'.

The paths of the accepted videos will be stored in a .txt file (default: accepted_videos.txt).

• parent_dir (required):
  The path to the dataset directory containing your clips.

• output_file (default: ./accepted_videos.txt):
  This file stores the paths of the videos that have been accepted.

• fps (default: 100):
  Frames per second for processing the clips.

• minimum_frames (default: 3):
  The minimum frame threshold. Videos with fewer frames than this will be skipped.

• restart (optional):
  If set to True, the process will restart from scratch, reloading both the video paths and state.

• random_selection (optional):
  Specify the number of clips to select randomly.

When a video is accepted, its path is added to a text file containing the list of all the accepted videos. You have to execute the copy_clips.sh file to copy the accepted videos to a desired location.