YOLOv8 QAT

Original Ultralytics compatible. (You can pretrain your model befor QAT as original way using this repository)

Usage

Install editable package in your environment by pip install -e .

Using pytorch native quantization API

python qat_pytorch.py \
--model-config ${model_config_yaml_file} \
--pretrained-weight ${path_to_your_pretrained_weight} \
--data-config ${path_to_your_data_config_file}

Using pytorch_quantization package from nvidia

You need to install pytorch_quantization package

python qat_nvidia.py \
--model-config ${model_config_yaml_file} \
--pretrained-weight ${path_to_your_pretrained_weight} \
--data-config ${path_to_your_data_config_file}

TODO

• end-to-end export to TensorRT engine(when using pytorch_quantization)
• code refactoring
• find other ways to improve mAP after QAT

References

https://medium.com/@DeeperAndCheaper/quantization-yolov8-qat-x2-speed-up-on-your-jetson-orin-nano-2-how-to-achieve-the-best-qat-8077ac0a167b

https://pytorch.org/tutorials/advanced/static_quantization_tutorial.html#quantization-aware-training