MVIG-SJTU / AlphaPose

Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System
http://mvig.org/research/alphapose.html
Other
8.05k stars 1.98k forks source link

Extremely slow pose estimation and detection in 3D_inference. #1151

Open arnavkundalia opened 1 year ago

arnavkundalia commented 1 year ago

Hi! I am using the demo_3D_inference for the webcam detection and it is showing extremely slow times. det time: 4.2723 | pose time: 20.0649 | post processing: 0.2372

The GPU is hardly being used and the CPU is being the bottleneck. python main.py --webcam 0 --vis --profile --vis_fast --sp: I use this command. Thanks

Fshaw123 commented 1 year ago

hello , which kind is your GPU

arnavkundalia commented 1 year ago

Its a RTX 3060, Laptop GPU.

bmmtstb commented 1 year ago

Does torch have cuda access? e.g. run in environment: python -c 'import torch;print(torch.cuda.is_available())'

What is your cuda-toolkit and torch version?

sunwuzhou03 commented 1 year ago

hello, do you know how to use the alphapose to get the 3d pose data? I don't find the documents about the 3D. Thank you!