/TCNModels

Primary LanguagePython

Action segmentation using TCN models for pytorch

Personally refined code for Multi-Stage TCN (CVPR2019)

Datasets

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

Training

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

Evaluation

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

Features

・For extracting features we use the I3D model , which are usually used in action segmentation tasks

References

Multi-Stage TCN (CVPR2019)