Documentation on multiple media files of the same type with WebDataset

[DCNemesis]

The current documentation on a creating a video dataset includes only examples with one media file and one json. It would be useful to have examples where multiple files of the same type are included. For example, in a sign language dataset, you may have a base video and a video annotation of the extracted pose. According to the WebDataset documentation, this should be able to be done with period separated filenames. For example:

e39871fd9fd74f55.pose.mp4
e39871fd9fd74f55.json
f18b91585c4d3f3e.base.mp4
f18b91585c4d3f3e.pose.mp4
f18b91585c4d3f3e.json
... 

If you can confirm that this method of including multiple media files works with huggingface datasets and include an example in the documentation, I'd appreciate it.

[lhoestq]

Yes this is correct and it works with huggingface datasets as well ! Feel free to include an example here: https://github.com/huggingface/datasets/blob/main/docs/source/video_dataset.mdx