The package manager vcpkg was utilized to install OpenCV (4.10).
Installation:
! Ensure that the environmental variable VCPKG_ROOT points to your vcpkg installation for Cmake to pick it up properly.
- CPU ->
vcpkg install opencv4[dnn] - GPU, with CUDA support ->
vcpkg install opencv4[dnn-cuda]
Note: you may need to separately install CUDA and CuDNN from Nvidia, if not present on your system. - GPU, without CUDA support ->
vcpkg install opencv4[dnn,opencl]
Note: OPENCV_OCL4DNN_CONFIG_PATH tba
You can also explicitly tell vcpkg which platform to build for by adding the postfix :your-platform to the end of the command.
Regarding vcpkg's supported platforms, refer to here.
Me, working on Windows 64-Bit with CUDA, used vcpkg install opencv4[dnn-cuda]
Pro tip:
For sped up builds, we recommend setting the CMAKE_BUILD_PARALLEL_LEVEL environmental variable to the amount of logical processors (threads) your CPU supports, to ensure maximum parallelization.
On Windows, that can be done with the command set CMAKE_BUILD_PARALLEL_LEVEL={amount_max_threads}.
The YoloV7 repository is included in the /yolo/v7 folder, as a git submodule.
To have it output a model in the onnx (dnn runtime compatible) file format, run the following command:
python export.py --weights yolov7-tiny.pt --grid --end2end --simplify --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --max-wh 640
The YoloV9 repository is included in the /yolo/v9 folder, as a git submodule.
To build using it, you either need to download a pretrained model (e.g. via the ones pretrained on the COCO set via https://github.com/WongKinYiu/yolov9/?tab=readme-ov-file#performance)
s
To then convert the .pt file to .onnx via the ONNX pipeline:
python export.py --weights yolov9-m.pt --simplify --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --imgsz 640 640 --include onnx
(or python3 for unix systems)
The OC-Sort repository is included in the /oc-sort folder, as a git submodule.
Yolo v9m, OC-Sort & Cross-Camera matching: 43ms inference on 2x 640x640 camera input feeds
Wong Kin-Yiu, for his YoloV7 and YoloV9 implementations.
Jinkun Cao for OC-Sort.
Fernando B. Giannasi for his implementation of the Hungarian algorithm in C++.