Open egeozsoy opened 1 year ago
Actually, in the main paper, we used all the training set for training and test on the test set.
However, in the PSG competition that we held last year, we use psg_train_val.json
(45697 training data + 1000 validation data with GT) as training and val splits, so that the rest are test set. But again, we dont have validation set in the original paper.
Hi, what is the currently recommended train/val/test split?
The line
assert split in {'train', 'test'}
confuses me a bit, as it indicates that there is no validation split?