SVO recording for body tracking

Hello all,
I am using ZED2 mainly for body tracking applications. So far, I have been using the real-time tracking. Although I use the camera refresh rate of 60 fps, I only get ~7 data points per second (output of 7Hz).
Now, I was considering doing the body tracking from a recorded video in SVO.

  1. Is it possible to do the body tracking on a recorded SVO video? (I think it is possible but I just want to confirm)
  2. If I recorded the video in SVO and then used the body tracking feature, would that increase the data output? In other words, will I get a higher number of data points if I use SVO? or is it going to be the same?

Note: I have a good Nvidia GPU with a lot of VRAM, so my hardware should be more than capable of handling the computational load.

All your help is much appreciated,

Hi @feraskiki,

You can totally do body tracking from an SVO. If you use our sample, you can just pass the path to your SVO as an argument when launching the program.
I don’t think you should get a better output by using an SVO itself.

The performance are impacted mostly by the depth mode and and the body tracking model used. You could get lower fps by using Neural depth mode along human body accurate for instance.

That being said, 7 new bodies input / seconds sounds low. What is your exact hardware, if I can ask?

Thank you for your support.
Actually, I tried it on multiple devices. I have gtx 1050 and rtx 30 and results were similar in both 3.8 and 4.0 sdk.
What is the number of data points/ seconds that you are getting or one is expected to get?

On a 30X0, you should get 30 fps no problem. On the 1050, it will be probably way less.
How many fps / depth frequency do you have on Depth Viewer?