are the GTX cards capable of isolating whole bodies from more than one live video in TD (run multiple instances of the TOP "Nvidia Background)? A resolution of 720p for each live-video would still be acceptable.
a litte background (no pun intended ): For a room installation i need to realtime-key (aka backgroundremoval) visitors in four live videos. My test with a kinect (v2) went well (playerID>blur & RGB > matte > out), but unfortunately I am not satisfied with the resolution of the kinects playerID stream.
I was trying to stay on budget this way, because I could have been working with four older (but still performant barebone computers, but without too fancy GPUs) and four relatively cheap kinect V2s.
If the GTX cards and TD can only key (aka backgroundremoval) ONE live video each and I thus would need four computers with such cards, it breaks my budget
Just for clarity, I don’t believe the Background TOP will work on any GTX cards, but it should run fine on the RTX 2060 or 3060 cards you mention. I also have a higher end card in my system right now, so I can’t confirm the minimum for 4 streams, but generally speaking it doesn’t take a lot of resources and there are 2 quality levels that can be used to adjust performance.
Another difference, the Nvidia Background TOP only works for one person in the frame where as the Kinect can isolate multiple people. The Kinect Azure has higher resolution and can detect multiple people in frame but you might need a dedicated USB card to handle the bandwidth of running 4 cameras on one machine.
From nvidia’s documentation (AI Green Screen is what Nvidia calls the engine that is used in the background top):
The AI green screen effect achieves its best results on videos that are recorded by one person sitting in front of a camera. The feature will not perform well on full-body videos, multiple persons in the scene, or camera angles that deviate too much from a front-facing camera.
excellent, thank you all! To sum it up: Multiple instances of the TOP are okay, but full body, more than one person in frame and freely moving bodies are not suggested by nvidia.