Hi, I checked out the paper and the testing code and found it is an interesting work.
But there is no training data to reproduce the same pretrained models.
I wonder if you have any plan to release the training data including static images and video lists(as described in the paper)?