Skeleton based multi-person behavior detection

[nlxxWangYH]

I want to perform spatiotemporal behavior detection (multiple behaviors) on multiple people in the video at the same time, I check the data preparation chapter of the official document, where the dataset for spatiotemporal behavior detection is the AVA dataset, but the dataset does not use skeleton information. In addition, I also noticed that there is a dataset processing part dedicated to behavior recognition based on the skeleton in the official document, and after I checked it, I found that the number of people in its use scenario did not exceed 2, and it was mainly used for behavior recognition and not spatiotemporal behavior detection.
How do I structure my dataset if I want to detect spatiotemporal behavior for multiple people in a video based on skeleton information?
Thank you very much for checking out my problem in your busy schedule, and I hope you can help me solve this simple problem.

[cir7]

Currently, MMAction2 doesn't support training skeleton-based spatio-temporal aciton detection models. We provide a simple inference demo, maybe you could refer to it

[nlxxWangYH]

Thank you for your answer.
Is there any future plan for mmaction2 to support spatiotemporal behavior detection methods based on skeleton.

[cir7]

yes, we have a plan to support it, but can't guarantee a specific time.

[nlxxWangYH]

Absolutely good news, thank you for your contributions and look forward to good news in the future.

[SCsleepy]

@nlxxWangYH
May I ask if you have implemented it? Can you teach me. Thank you very much.