Felt like trying to optimize a yolo model as much as I could running solely on CPU. Trained a Python CNN on the fruits360 dataset with lots and lots of augmented data.
Ended up spening a couple hours on this, optimized with onnx and int-8 quantinization and got around ~75fps. I think I could of optimized heavier but this was enough given this was more to scratch curiosity.
I also know that the video below is NOT 75 fps, recorded it before optimizations but the tomato did not survive the making of this video and have not been bothered to re-record.