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 315 forks source link

TRT8 on Windows #141

Open adventurousbanana opened 3 years ago

adventurousbanana commented 3 years ago

How can this be built with TensorRT 8 and Cuda 11.3 on Windows?

I tried building the TRT8 branch for use with yolov5s, but I encountered several errors. Is the branch up to date?

Any help would be greatly appreciated.

seungtaek94 commented 3 years ago

https://github.com/enazoe/yolo-tensorrt/issues/51#issuecomment-917840303

I also have issue in cuda 11.3 and tensorrt 8 :(

enazoe commented 3 years ago

@seungtaek94 past the error log, I have test the trt8 branch code on windows (vs2017 cuda10.1 trt8.0.1.6 gtx1080ti) everything is all right.

seungtaek94 commented 3 years ago

@seungtaek94 past the error log, I have test the trt8 branch code on windows (vs2017 cuda10.1 trt8.0.1.6 gtx1080ti) everything is all right.

What should value that _n_size_split be ?

https://github.com/enazoe/yolo-tensorrt/blob/cc405b07af4351334e7b5f47dc70bec73ab25a06/modules/chunk.cu#L68

image

I just try other model yolov5s from this repo, Actually it works..

Before I used yolov4-tiny.cfg, yolov4-tiny.weights from AlexeyAB/darknet.

adventurousbanana commented 3 years ago

@seungtaek94 past the error log, I have test the trt8 branch code on windows (vs2017 cuda10.1 trt8.0.1.6 gtx1080ti) everything is all right.

What should value that _n_size_split be ?

https://github.com/enazoe/yolo-tensorrt/blob/cc405b07af4351334e7b5f47dc70bec73ab25a06/modules/chunk.cu#L68

image

I just try other model yolov5s from this repo, Actually it works..

Before I used yolov4-tiny.cfg, yolov4-tiny.weights from AlexeyAB/darknet.

You got it build and run with yolov5s? Did you just change the include directories and additional dependencies for OpenCV, Cuda, and TensorRT in Visual Studio 2017?

seungtaek94 commented 3 years ago

Sorry for late response.

I didnt change anything in code. I chaged only *.cfg and *.weights path for each model.

Actually, I tested yolov4-csp.cfg, yolov4-tiny.cfg and yolob5s. Only yolov4-tiny.cfg didn't work, others worked.

Nuzhny007 commented 2 years ago

https://github.com/enazoe/yolo-tensorrt/issues/148

And second problem: _n_size_split is uninitialized. Write here inplace initialization: https://github.com/enazoe/yolo-tensorrt/blob/TRT8/modules/chunk.h#L61

int _n_size_split = 0;