@yrcong What train-val-test split is used with Action Genome dataset? From the codebase, it seems like the same test set was used as validation set during training (in train.py) and also as test set in (test.py). What dataset split is used for reported metrics in paper?
@yrcong What train-val-test split is used with Action Genome dataset? From the codebase, it seems like the same test set was used as validation set during training (in train.py) and also as test set in (test.py). What dataset split is used for reported metrics in paper?