Installation

pip install -r requirement.txt

Make sure to download ffmpeg on your device

Import Video Data

python videos/get_videos.py

After importing video data completes, you can then extract different actions with the command below

python data_preparation/create_action.py --input "Brief screenning.MOV" --action "moving backwards" --start "00:00:39" --end "00:00:49"

Where --input should be a video already in the videos folder --action should be the action discovered in the input video --start is the time the action begins --end is the time the action ends

Building Actions Dataset

python data_preparation/action_dataset.py

Train Actions Dataset

python train_slowfast.py

You'd get results like this

# 500 epochs

# Testing DataLoader 0: 100%|███████████████████████████████████████████████████████████████████████████| 3/3 [00:12<00:00,  0.23it/s]
# ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
#        Test metric             DataLoader 0
# ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
#         test_acc             0.800000011920929
#         test_loss           0.6259332895278931
# ──────────────────────────────────────────────────────────────

Test Action Dataset

video_path = 'videos/Postural Analysis Lateral.mp4'
all_actions = generate_actions_from_video(video_path)
print(all_actions)

python3 test_slowfast.py

tensorboard serve --logdir lightning_logs/