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

YOLOv5 Segmentation Fault #108

Closed peterchondro closed 3 years ago

peterchondro commented 3 years ago

Hi,

I am trying to convert YOLOv5x with customized model (nc = 4, width = 416, height = 256). Conversions from yaml to cfg and pt to weights were successful. But when I tried to generate TRT engine, it will result in a segmentation fault.

I also tried to convert YOLOv5-3.0 (x) and YOLOv5-4.0 (x) (cfg and weights) from your repo; YOLOv5-3.0 (x) worked fine but YOLOv5-4.0 (x) generated a segmentation fault.

PS: I found my YOLOv5x and your YOLOv5-4.0 cfgs have C3 layer, which was not known in my side. So I have to change these to BottleneckCSP. I am not sure if it can help with this issue.

peterchondro commented 3 years ago

Fixed with new yolo.cpp by author.