The sum of number of videos in train_split.txt and val_split.txt is greater than the total number of videos in the dataset. Is this a bug?

[ut21]

There are total 3275 videos in train_split.txt, and 1402 in val_split.txt while the total number of videos I was able to extract (by stitching together the frames) was 3993, while the paper reports total 4677 videos but had empty frames for multiple videos (as is pointed out by currently open issues). Moreover, there are 848 videos which are mentioned in either the train_split.txt or val_split.txt but are not present in the extracted videos, which means 684 videos are missing from the dataset. And there are 164 videos that were extracted which aren't present in neither train_split.txt nor val_split.txt, what are these extra videos? Thanks, any clarification or updates on this will be greatly appreciated as we want to use this dataset for our experiments