Daniil-Osokin / lightweight-human-pose-estimation-3d-demo.pytorch

Real-time 3D multi-person pose estimation demo in PyTorch. OpenVINO backend can be used for fast inference on CPU.
Apache License 2.0
653 stars 137 forks source link

On my machine, fast pose extractor is working very slow #97

Closed mix0z closed 2 years ago

mix0z commented 2 years ago

Fast pose extractor is working about several seconds on each image, in the same time legacy slow implementation works less then second on each image. What is the reason?

Daniil-Osokin commented 2 years ago

Hi! This looks strange, possibly you built it in a Debug mode or linked against libraries in a Debug mode. Also how did you measure the time? Possibly your system started a background task at that time.

mix0z commented 2 years ago

No, I didn't built in Debug mode. I mesured time, using datetime.now(), before and after each call of exctract_poses().

Also some time ago, fast pose extractor worked nice. But after smth changed in the system maybe (I don't exactly now what), it began to work slow. I didin't changed code of pose extraction.

Daniil-Osokin commented 2 years ago

Ok, closing.