search

Megvii-BaseDetection / YOLOX

YOLOX is a high-performance anchor-free YOLO, exceeding yolov3~v5 with MegEngine, ONNX, TensorRT, ncnn, and OpenVINO supported. Documentation: https://yolox.readthedocs.io/
Apache License 2.0
9.48k stars 2.22k forks source link

replace Focus with Conv #998

Open ljmiao opened 2 years ago

ljmiao commented 2 years ago

It is recommended to replace Focus with conv, because Focus is extremely unfriendly to deployment, and doing so has almost no negative impact on the model, which has been confirmed in yolov5. link: https://github.com/ultralytics/yolov5/issues/4825

FateScript commented 2 years ago

Thanks, I will have try. I will inform you If anything updates.

FateScript commented 2 years ago

Using kernel_size 6x6 in Conv2d is not a good way to replace Focus. Anyway, since so many users are willing to replace Focus with Conv2d, we will try to do such a thing in a different way.

scsolaris commented 2 years ago

Using kernel_size 6x6 in Conv2d is not a good way to replace Focus. Anyway, since so many users are willing to replace Focus with Conv2d, we will try to do such a thing in a different way.

@FateScript any progress?