CAIC-AD / YOLOPv2

YOLOPv2: Better, Faster, Stronger for Panoptic driving Perception

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

GPU performance issues

DexterTV opened this issue · comments

Hi,

I tried a modified demo.py as recommended in the issues using the link https://debuggercafe.com/inference-using-yolopv2-pytorch/ but I don't get the improvement of fps switching from cpu to gpu.
I used the pre-trained model on an NVIDIA A4000 GPU and as input a 482x302 jpg image and updated pytorch to 1.13.1.

Result fps with cpu: 2.5
Result fps with gpu: 0.6

Does anyone know if I'm doing something wrong?
Thank you in advance.

Update: I tried Google Colab enabling GPU (NVIDIA Tesla T4) with torch version 1.13.0+cu116 and inferred an image 100 times and got 57 of average fps instead with NVIDIA A4000 and updated torch the performance is very low (see previous comment).
What could be the problem?