config.yaml

# root directory of your dataset
data_root: 'data/demo/videos'

# annotation files of your dataset, you need to organize them following demo list
ann_file_train: 'data/demo/annotations/train_list.txt'
ann_file_val: 'data/demo/annotations/val_list.txt'
ann_file_test: 'data/demo/annotations/test_list.txt'

# video length of your dataset (second),
# if your video length is inconsistent or you don't know the specific length of the video, you can set it to the average length or directly fix it to 2 seconds
video_length: 2

# number of classes
num_class: 25

# batch size while training, it depends on your gpu memory
videos_per_gpu: 8

# total epoch
epoch: 50

# working directory during training, we will look for the checkpoints file in the same directory while testing
work_dir: 'work_dirs/demo'

# output path of test result
test_results_path: 'results.json'