Nathan Louis

It is the **Hands with Manual Keypoint Annotations** dataset from [here](http://domedb.perception.cs.cmu.edu/handdb.html).

The cropping is done in the dataset file: https://github.com/MichiganCOG/ViP/blob/master/datasets/Manual_Hands.py#L82 It's more efficient to make those changes on-the-fly rather than creating new images.

Yeah collate is very useful for zero-padding your samples when there's a different number of objects (or lengths in other cases). But in my experience, it's specific to each problem....

I saw this change in your pull request #30, but it wasn't backwards compatible.

That should've been fixed with issue #7 with the following line: https://github.com/MichiganCOG/ViP/blob/dev/train.py#L182. Do you have this version from dev pulled?

That directory should be updated to point to your copy of the HMDB51 dataset ($JSON_DIRECTORY). Please use this file: https://github.com/MichiganCOG/ViP/blob/master/datasets/scripts/gen_json_HMDB51.py to generate that file. Note that you must edit the...

We have now made all raw video files available for download for non-commercial, research purposes only. They can be found on the YouCook2 [Downloads](http://youcook2.eecs.umich.edu/download) page.

Hi, unfortunately we no longer have access to update those links. However, many of those pre-trained weights can be found from other existing repositories.