Question about training procedure when dataset is devided with train/test set.

[jacob-kang]

Hi.
Thanks for your great work!

I recently studied about video domain.

And I wonder the training procedure.

I figured out that UCF-101 dataset has train and test set.

But Video generation not prediction, Train and Test set are meaning less.
Because It is just generation. So there is not ground truth but can only measure by using FVD.

In that case with train/test set dataset (like UCF-101), Do you use train set only when you train your model? or use both of them?

Thanks.

kind regards.