SthPhoenix / InsightFace-REST

InsightFace REST API for easy deployment of face recognition services with TensorRT in Docker.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

why did you prefer to use cython than numba gpu?

ThiagoMateo opened this issue · comments

I'm not using Cython anywhere beside tests right now.
Actually Numba is actively used, but it's CPU version.

Numba can't be used with arbitrary python code and still doesn't support some numpy functions, so it can be used to speed up some parts of code, and it's working great with NMS right now.

As for GPU version - I haven't tested it yet, but I think it might cause additional overhead on pushing/pulling data to/from GPU between parts of code which can't be wrapped with Numba.

Closing, as it seems to be resolved )