search

enazoe / yolo-tensorrt

TensorRT8.Support Yolov5n,s,m,l,x .darknet -> tensorrt. Yolov4 Yolov3 use raw darknet *.weights and *.cfg fils. If the wrapper is useful to you,please Star it.
MIT License
1.19k stars 316 forks source link

yolov3-voc #134

Open zhangjian-neu opened 2 years ago

zhangjian-neu commented 2 years ago

There is precision lost !!! It is very obvious for yolov3-voc.cfg. Big gap was found between this implement and the original darknet implement for the picture "data/person.jpg" in darknet package.

enazoe commented 2 years ago

精度损失是正常的，trt量化之后是会有变化的

zhangjian-neu commented 2 years ago

感谢您的回复。

我现在有点怀疑的是：本程序对标准yolov3.cfg做了特定处理，不适配yolov3-voc.cfg。yolov3-voc.cfg内的配置参数於yolov3.cfg略有不同，而且训练时使用的是VOC而不是COCO数据。这两点有影响吗？

请不吝赐教！