Hello, when I used the same onnx model to detect the original yolov5 project and this project, I encountered the problem of different results. The original yolov5 project category result is correct, but only part of the category of this project is correct, the recognition box The same, but the confidence level is also different. How to solve this problem?

Hi, I have built onnxruntime on macOS, but under MacOS/Release folder, there is no onnxruntime_cxx_api such file at all. Also, there is no such lib sub folder under it, it just inside MacOS/Release/libonnxruntime.dylib.

How should I set them under macOS?

just got 2 such file in src:

/libs/onnxruntime//cmake/external/onnxruntime-extensions/includes/onnxruntime/onnxruntime_cxx_api.h
/libs/onnxruntime//include/onnxruntime/core/session/onnxruntime_cxx_api.h

Hello, I had an experiment where all the images were high pixels, so scaling to 640*640 would cause the target to be too small. I tried to modify the c++ file in the src folder to change 640 to 1280, but after compiling, I still need 640 input, so how should I modify the project?

Hello, I am trying to run this example, but when I'm writing "cmake --build ." in the terminal, I always get this error:

LINK : fatal error LNK1104: Datei "onnxruntime-win-x64\onnxruntime-win-x64\lib\onnxruntime.lib.lib" kann nicht geöffnet werden. [...\build\yolo_ort.vcxproj]

Maybe the "lib.lib" from onnxruntime.lib.lib is the problem, but I don't know how to solve it. I am using the win-x64-1.10.0 onnxruntime version.

Thanks for your help.

There seem to be an issue handling an input other than 640x640. When I try to feed a 320x1296 input it throws an error: Got invalid dimensions for input: images for the following indices index: 2 Got: 1296 Expected: 640 index: 3 Got: 320 Expected: 640

I think it has to do with the dynamic input shape checking of the code, which I think it is not doing its job correctly. Can someone point me where should I look at, to make it able to excecute multiple input shape images? Thanks!

hello,I have tested ort c++ inference successfully! but I didn't make batch inference works! could you please give a batch inference c++ example?? Thank you very much!

Official yolov5 PyTorch repo uses half precision. I try the onnx model with half precision on python, and speed increased. Can this repo support half precision？

In preprocessing

void YOLODetector::preprocessing(cv::Mat &image, float*& blob, std::vector<int64_t>& inputTensorShape)
{
    cv::Mat resizedImage, floatImage;
    cv::cvtColor(image, resizedImage, cv::COLOR_BGR2RGB);
    utils::letterbox(image, resizedImage, this->inputImageShape,
                     cv::Scalar(114, 114, 114), this->isDynamicInputShape,
                     false, true, 32);

    inputTensorShape[2] = resizedImage.rows;
    inputTensorShape[3] = resizedImage.cols;

    resizedImage.convertTo(floatImage, CV_32FC3, 1 / 255.0);
    blob = new float[floatImage.cols * floatImage.rows * floatImage.channels()];
    cv::Size floatImageSize {floatImage.cols, floatImage.rows};

    // hwc -> chw
    std::vector<cv::Mat> chw(floatImage.channels());
    for (int i = 0; i < floatImage.channels(); ++i)
    {
        chw[i] = cv::Mat(floatImageSize, CV_32FC1, blob + i * floatImageSize.width * floatImageSize.height);
    }
    cv::split(floatImage, chw);
}

env: platform: windows 10 x64 onnxruntime version: gpu 1.7.0

when processing excute to AppendExecutionProvider_CUDA function, occur an exception, and i found cudaOption obeject Member variables not right, for exanmple the device_id is a Uninitialized values like -858993460, can you help me ? thank you very much!

This is great reference for c++. Question: In Line https://github.com/itsnine/yolov5-onnxruntime/blob/master/src/detector.cpp#L112 . Why are we considering only first element of outputTensors, It has 4 output array. We could request for all 4 outputs if we change parameters in https://github.com/itsnine/yolov5-onnxruntime/blob/master/src/detector.cpp#L190

Any particular reason to go this way?

I could not find any reference including Anchor_boxes, Could you please add one? Thanks.

First of all, thanks for the working C++ code that uses onnxruntime with yolov5 :+1: Could you please add the license file to the repository so that it is clear how your code can be used in other projects

I tried out your sample - very cool! I get 110 FPS with a YOLOv5s running CUDA 11.5 on my 1080ti. I am curious what it would take to evaluate performance with TensorRT. Have you tried this? Any pointers? Thanks.