Yolo on edge CPU

Show and Tell
1

Hey!

I have spent a decent amount of time attempting to build a yolo model specifically for CPU optimization, the goal here was to build a easy to use and straightforward model family with heavy focus on edge devices. This was originally never meant to be shared but after extensive testing I can’t help to feel that it would be a shame not to share it. For context the last testing was done with —p2 and image size 320 I managed to retain the accuracy from the 640 benchmarks but got an insane 200 fps (model inference) on my desktop CPU. If this can help anyone out AWESOME! If you find a bug or wish to contribute just leave a comment.

Disclaimer!! This was vibe coded, even if the concept and ideas came from me ChatGPT has been used as a working horse, I’m simply and enthusiast who loves this kind of stuff.

Repo with benchmark: YoloLite-Official-Repo/BENCHMARK.md at main · Lillthorin/YoloLite-Official-Repo · GitHub

1 Like
2

Created two model cards for the models, one for 640x640 onnx inference and one for 320x320 with p2 head activated! Feel free to test the models out.

1 Like
3

Way to go, keep up the good work!

2 Likes