Python scripts performing object detection using the YOLOv6 model in ONNX.
- About the YOLOv6 name: https://github.com/meituan/YOLOv6/blob/main/docs/About_naming_yolov6.md
- The input images are directly resized to match the input size of the model. I skipped adding the pad to the input image, it might affect the accuracy of the model if the input image has a different aspect ratio compared to the input size of the model. Always try to get an input size with a ratio close to the input images you will use.
- The original YOLOv6 repository is still in development, so things might not work in the future.
- Check the requirements.txt file.
- For ONNX, if you have a NVIDIA GPU, then install the onnxruntime-gpu, otherwise use the onnxruntime library.
- Additionally, pafy and youtube-dl are required for youtube video inference.
git clone https://github.com/brain-facens/YOLOv6
cd ONNX-YOLOv6-Object-Detection
pip install -r requirements.txt
For Nvidia GPU computers:
pip install onnxruntime-gpu
Otherwise:
pip install onnxruntime
pip install youtube_dl
pip install git+https://github.com/zizo-pro/pafy@b8976f22c19e4ab5515cacbfae0a3970370c102b
The original model was converted to ONNX using the following Colab notebook from the original repository, run the notebook and save the download model into the models folder:
- Convert YOLOv6 ONNX for Inference
- The License of the models is GPL-3.0 license: License
The original Pytorch model can be found in this repository: YOLOv6 Repository
- Runing:
./run.sh
- Comparison with YOLOv5:
python video_object_detection.py
- YOLOv6 model: https://github.com/meituan/YOLOv6
- YOLOv5 model: https://github.com/ultralytics/yolov5