Closed blazapropeler closed 1 year ago
We have not tested the speed, but from my experience the inference is very fast with GPU. It seems like real-time works from this issue.
I have tested the pose3d, it was extremly fast. However I found the mesh3d model ran twice in infer_wild_mesh file. @Walter0807 Do u know how much effect loose if we just using single forward on mesh3d inference?
I have tested the pose3d, it was extremly fast. However I found the mesh3d model ran twice in infer_wild_mesh file. @Walter0807 Do u know how much effect loose if we just using single forward on mesh3d inference?
You can remove the flipping part if speed is important to you, it would slightly increase the error from my experience. You might refer to this script.
hello @lucasjinreal, I recently try to inference motionbert in real time, but i'm stuck. Now I can use AlphaPose to do 2D extraction in real time, however, i don't know how to use this real time 2D extraction as an input to the motionbert, can you guide me or give me some advice? thank you so much!
@XALEX-123 Actually I found motionBert has large performance reduce if you are not sampling on whole sequence, you should try on python first, take a deep look at performance in realtime situation.
@lucasjinreal thank your rapid reply, then how can i inference motionbert in real time?
@lucasjinreal can your share your main env and GPU on your PC, how many fps ?
Hi!
I was just wondering if you have some results on the speed and if this model (in the Lite variant) would be suitable for a real-time 3d pose estimation problem?
Thanks