Steps to Run 👋

First download YouCook2 data from http://youcook2.eecs.umich.edu/ and place them in folder named feat_csv.

Then run make_feat.py for both training and testing/validation files to create the feature files as .npy dump.

Run python3 siam-correct-cook-dtw-comment.py to start training the UNHCLE model. Models are saved as .pth extensions which can be easily saved and loaded using torch.load().

The model is trained for 100 epochs.

For evaluation, set batch_size = 1 and use appropriate feature files in dataloader and training code.