Haodong Duan

We are not familiar with that, either.

> @kennymckormick I think if I use my own video set to train n classes, what configuration needs to be changed? thanks！！ Actually, lots of codes need to be modified...

Have you already downloaded the Kinetics videos and got frames extracted? which are steps 2 & 3 in the tutorial

OK, maybe then you need to first check the downloading scripts.

It is possible. You just need to parse the AST of the config, and then change the node value and write it back to the config. You can not directly...

Thank you for your attention, we will notify you once we released the RGBPose-SlowFast part.

GradCAM hasn't supported the transformer-based models yet. @irvingzhang0512 do you have some time to look at this problem?

Sure, these two tutorials are on our schedule

> Can you give an estimated date? Tutorials for skeleton-based action recognition will be released in a week. For spatiotemporal action detection, it will be released in August.

Do you mean the AUC / AR@N for training videos?