hexmagic / onnx-yolov5 Goto Github PK

Hi, I built the main.exe using one computer with RTX 3060, and I copied the debug main.exe, yolov5s.onnx, opencv_world455d.dll, coco.names, test.jpg to my laptop with GTX 950m.

I have built the main.exe successfully, and run the main yolov5s.onnx test.jpg successfully. However, I found it is using the CPU of my laptop, not the GPU, while I do have a GPU GTX 950m.

What am I missing?

seems postprocess is not right?
https://github.com/Hexmagic/ONNX-yolov5/blob/master/src/test.cpp#L49

you just handle one output

How to implement on Windows system？When I want to assign my own model path, I will fail to read the onnx model.

Hi，
I was trying to generate YOLOV5 onnx model with
python export.py --weights yolov5s.pt --include onnx
But, it turns out to be
ONNX: export failure ❌ 0.0s: Unsupported ONNX opset version: 17
So why the onnx cannot be exported with the default opset version 17?

Hi,

Thanks for sharing your code. I have a Pytorch models (*.pt) trained with custom data, and have them converted to onnx using export.py from Ultralytic's yolov5. ONNX-yolov5 works on Windows 10 as expected. Then have ONNX-yolov5 compiled and built on uBuntue 22.04 and ran it with the same onnx models, but it output incorrect bounding boxes. I am wondering if there is I have pt files converted to onnx files corrected. When running export.py I have tried with or without --simplify as suggested from one of posts on the internet. Can you suggest right tools for converting pt files to onnx files? Thanks,

CC

Hi,

Is it possible to support the classification task?

Thanks

i have had a problem with building so have to change CMakeLists.txt to:

cmake_minimum_required(VERSION 3.0 )
project(app)
set(CMAKE_CXX_STANDARD 14)
set(CMAKE_CXX_STANDARD_REQUIRED True)
SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread")
find_package(OpenCV REQUIRED)
include_directories(
    ${OpenCV_INCLUDE_DIRS}
    ${PROJECT_SOURCE_DIR}/include
)
add_executable(main main.cpp  src/detector.cpp src/loguru.cpp)
target_link_libraries(main  ${OpenCV_LIBS} ${CMAKE_DL_LIBS})
add_executable(test src/test.cpp)
target_link_libraries(test  ${OpenCV_LIBS} ${CMAKE_DL_LIBS})

and when converting in original yolov5 dir:
python export.py --weights yolov5s.pt --include onnx --opset 12

after that should work:
./build/main yolov5s.onnx data/images/zidane.jpg

Why do model loads always report errors?The model path has also been modified.

Why does “dnn::readNetFromONNX("./best.onnx")”dnn::readNetFromONNX("./best.onnx") report an error when I run test.cpp?

https://github.com/Hexmagic/ONNX-yolov5/blob/master/src/test.cpp#L73
As you see, indices[ind] will cause a crash.

I need help🤕

i see inference time about 1,7 sec with yolov5 onnx using
./build/main yolov5s.onnx data/images/bus.jpg

almost the same inference time (1,3 sec) i see with yolov5 onnx:

there is no advantage ? or i missed something ?

hello,when I build the net by readNetFromONNX(), I get this error? How to solve it? Thank you!

Hi,
I have a model trained with an input img size of 416 and a batch size of 16.
how should the code change in this case?
thank you
( I had the following error )

Hi, does this code only support batchsize=1 and if I want to change the batchsize, how do I start?