Personally refined code for Multi-Stage TCN (CVPR2019)
Need to download the dataset before starting. Download the data directory from the link which is in the original code and place it in the top directory level, same as main.py.
Data consists 3 datasets
・GTEA , 50salads , breakfast
python main.py [arguements(e.g. template.yaml)] train [split(1-4 for breakfast and gtea, 1-5 for 50salads)] --[device(e.g. cuda:0)]
check the template.yaml for example of settings
python main.py [arguements(e.g. template.yaml)] test [split(1-4 for breakfast and gtea, 1-5 for 50salads)] --[device(e.g. cuda:0)]
to output the predictions in the result directory
then
python eval.py [arguements(e.g. template.yaml)] [split(1-4 for breakfast and gtea, 1-5 for 50salads)]
to show the evaluation results
・For extracting features we use the I3D model , which are usually used in action segmentation tasks