[Feature] How can I use a skeleton based model trained with my dataset to predict real time actions with video streams from RTSP?

[PhucDucMai]

What is the problem this feature will solve?

Currently, inferring actions is based on a video to judge actions. How can I do the following: Read the video stream from the RTSP Link, create a frame buffer, and when there are enough frames, I will generate an action label (Unlike today, putting in a video, whether long or short, only generates 1 action). I hope to hear from you soon and thanks for your help. Best wishes

What is the feature?

Real-time inference when a sufficient number of frames is available

What alternatives have you considered?

No response