yolov7_opencv_cpp

Object Detection using YOLOv7 and OpenCV DNN C++

Get repo and install dependences

git clone --recursive https://github.com/majnas/yolov7_opencv_cpp.git
cd yolov7_opencv_cpp/yolov7

# Install dependencies.
pip install -r requirements.txt
pip install onnx

Prepare custom weight

Download my custom yolov7 face detection using this cmd

cd cfg/training/
wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1INiC_M_ttd8xMpZ9CuSA1FTqUxZT4e1y' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1INiC_M_ttd8xMpZ9CuSA1FTqUxZT4e1y" -O custom_weight.pt && rm -rf /tmp/cookies.txt

Or place use your own custom yolov7 weight in following folder.

└── yolov7
    ├── cfg
    │   ├── baseline
    │   ├── deploy
    │   └── training
    │       ├── custom_weight.pt           <----- Place the custom weight here
    │       ├── yolov7-d6.yaml
    │       ├── yolov7-e6e.yaml
    │       ├── yolov7-e6.yaml
    │       ├── yolov7-tiny.yaml
    │       ├── yolov7-w6.yaml
    │       ├── yolov7x.yaml
    │       └── yolov7.yaml

Make a copy of yolov7/cfg/deploy/yolov7.yml and rename to yolov7_custom_weight.yaml then change number of class in line number 2 (nc=1). For my custom weight there is only one class (face).

cd ..
cp deploy/yolov7.yaml deploy/yolov7_custom_weight.yaml
# edit yolov7_custom_weight.yaml => Set nc in line 2 => nc=1 in my case which I have only one class

Moving reparameterization_yolov7.py to yolov7 directory.

# In root of repo directory
mv reparameterization_yolov7.py ./yolov7/reparameterization_yolov7.py
# Edit./yolov7/reparameterization_yolov7.py  
# 1 - Set number of classes
# nc=1 # change this base on number of classes in your custom model
# 2 - Set device
# device = select_device('0', batch_size=1) # if using GPU
# device = select_device('cpu', batch_size=1) # if using CPU

Reparameterization of model

cd ./yolov7
python reparameterization_yolov7.py

⚠️ **If you are using different version of yolov7 (yolov7x, yolov7-tiny, ...) use different reparameterizatioin script from here **: Be very careful here!

This will create another model (custom_weight_reparameterized.pt) in cfg/deploy/custom_weight_reparameterized.pt, which is reparameterized version of custom weight.

└── yolov7
    ├── cfg
    │   ├── baseline
    │   ├── deploy
    │   │   ├── custom_weight_reparameterized.pt   <------------- Reparameterized custom weight 
    │   │   ├── yolov7_custom_weight.yaml    
    │   │   ├── yolov7-d6.yaml
    │   │   ├── yolov7-e6e.yaml
    │   │   ├── yolov7-e6.yaml
    │   │   ├── yolov7-tiny-silu.yaml
    │   │   ├── yolov7-tiny.yaml
    │   │   ├── yolov7-w6.yaml
    │   │   ├── yolov7x.yaml
    │   │   └── yolov7.yaml
    │   └── training
    │       ├── custom_weight.pt                  <------------- Custom weight
    │       ├── yolov7-d6.yaml
    │       ├── yolov7-e6e.yaml
    │       ├── yolov7-e6.yaml
    │       ├── yolov7-tiny.yaml
    │       ├── yolov7-w6.yaml
    │       ├── yolov7x.yaml
    │       └── yolov7.yaml

Export to ONNX

To export ONNX we have to checkout to u5 branch, and export reparameterized version of custom weight to onnx and torchscript, to do this

git checkout u5
python export.py --weights cfg/deploy/custom_weight_reparameterized.pt --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640

Now we will have onnx and tochscript version of our custom_weight.pt

└── yolov7
    ├── cfg
    │   ├── deploy
    │   │   ├── custom_weight_reparameterized.onnx           <------------- onnx version
    │   │   ├── custom_weight_reparameterized.pt             <------------- Reparameterized custom weight
    │   │   └── custom_weight_reparameterized.torchscript    <------------- torchscript version version
    │   └── training
    │       └── custom_weight.pt             <------------- Custom weight

To compile and run cpp version

# In root of repo directory
cd cpp
mkdir build
cd build
cmake ..
make
./app "../../data/me.jpeg" "../../yolov7/cfg/deploy/custom_weight_reparameterized.onnx" 640 640

ℹ️ If you got following error, you must install opencv on your system.

CMake Error at CMakeLists.txt:7 (find_package):
  By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "OpenCV", but
  CMake did not find one.

  Could not find a package configuration file provided by "OpenCV" with any
  of the following names:

    OpenCVConfig.cmake
    opencv-config.cmake

  Add the installation prefix of "OpenCV" to CMAKE_PREFIX_PATH or set
  "OpenCV_DIR" to a directory containing one of the above files.  If "OpenCV"
  provides a separate development package or SDK, be sure it has been
  installed.

Figure 1: cpp prediction for me_cpp_pred.png