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
9.44k stars 2.21k forks source link

The provided param and bin in YOLOX-Android-ncnn seem to be converted from untrained weights. #757

Closed wangziren1 closed 3 years ago

wangziren1 commented 3 years ago

Thank you for the wonderful work. I use provided param and bin, and make a test. The result is very poor. Then I convert yolox_tiny.pth to param and bin, and the results seems good. I think you uploaded the untrained param and bin mistakenly. Please upload the right param and bin, because if someone is a novice, he or she wil be very confused.

In addition, the param converted from .onnx need to modified by hand because of the unsupported slice op of ncnn. I think you may provide some explanations and link for how to modify. Thank you again for the project.

wangziren1 commented 3 years ago

@FateScript

FateScript commented 3 years ago

Sorry for not noticing. Plz refer to this tutorial.

wangziren1 commented 3 years ago

ok.