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 315 forks source link

about yolov4 problem #172

Open tuteming opened 2 years ago

tuteming commented 2 years ago

custom yolov4 主要在3個 [yolov]層定義classes數,在它的前一層set filters=3(5+classes) 我想問的是yolov4的變型[Gaussian_yolo]層,在它的前一層set filters=3(9+classes) 我知道tensorrt 沒有[Gaussian_yolo],我是否可以把[Gaussian_yolo]改名為[yolov] 然後在yolo.cpp及yolov4.cpp中把相關的5改成9. yolo.cpp line 290,692,845,859,1149,1245 yolov4.cpp line 30,32,34,36,39,46(?) 依然用 yolov4來產生engine.這樣行的通嗎? 如果你認為可以,上面註記的行數有沒有問題?

謝謝 煩請賜教

tuteming commented 2 years ago

這是行不通的 抱歉