shubham-goel/4D-Humans

PHALP tracker taking lot of CPU utilization

Closed this issue · 1 comments

9e0 commented

Hi,

I find that PHALP tracker taking lot of CPU utilization, as numpy does not support GPU.
Enabled code to run on CUDA, still the CPU utilization is heavy.

I feel the performance might be better if numpy array transactions can be moved to GPU

Is there a way I can make the PHALP tracking to run on GPU

Thanks in advance

We haven't optimized that aspect, so you might want to look into this, but it's possible that you could do better by moving all the operations to GPU.