UFC101 leaderboard: https://paperswithcode.com/sota/action-recognition-in-videos-on-ucf101?p=i3d-lstm-a-new-model-for-human-action
Referred paper: https://arxiv.org/pdf/1705.07750.pdf
// eg. train 'tea' Task
// choose one from all tasks:
// ['tea', 'cereals', 'coffee', 'friedegg', 'juice', 'milk', 'sandwich', 'scrambledegg', 'pancake', 'salat']
python train_tasks_by_conv_1_to_1.py 'tea'
python train_tasks_by_conv_3_to_1.py 'tea'
// test and get result csv in 'results/final/'
python test_tasks_by_models.py
// results data(final csv file)
./results/
- final
// trained model
./trained/
// utils directory
./utils/