search

DeepLabCut / DLC-inferencespeed-benchmark

A database of inference speed benchmark results on various platforms and architectures
https://deeplabcut.github.io/DLC-inferencespeed-benchmark/
MIT License
4 stars 4 forks source link

RTX3090 #6

Open wux024 opened 2 years ago

wux024 commented 2 years ago

I used RTX3090 for evaluation on Windows 10, and I was not sure if I got the correct results. I ran unpickle.py and got some errors.

D:\DLC-inferencespeed-benchmark\unpickle.py:15: RuntimeWarning: divide by zero encountered in true_divide results_fps = (1.0 / times for times in results) D:\Anaconda3\envs\dlclive\lib\site-packages\numpy\core_methods.py:202: RuntimeWarning: invalid value encountered in subtract x = asanyarray(arr - arrmean)

AlexEMG commented 2 years ago

Ok, I suppose your runtimes were 0?

AlexEMG commented 2 years ago

I suppose the culprit is, that if the code somehow crashes you can end up with 0 runtimes... https://github.com/DeepLabCut/DeepLabCut-live/blob/d8acac6420a5d1c1a31f8f87f46d0227c9cb734f/dlclive/benchmark.py#L289