YOLOV5和YOLOV8-加速

【YOLOv5】【模型压缩与加速】【量化】FP32、FP16、INT8_yolov5 export int8-CSDN博客

1. YOLOV5

FP32量化

python export.py --weights xxx.pt --include onnx engine --device 0
python detect.py --weights xxx.engine --source 0
# 量化前后：3.2ms/1.7ms

FP16量化

python export.py --weights xxx.pt --include onnx engine --half --device 0
python detect.py --weights xxx.engine --source 0
# 量化前后：3.2ms/0.8ms

2. YOLOV8

yolo task=detect mode=export format=engine model=weights/yolov8s.pt 
yolo task=detect mode=predict model=weights/yolov8s.engine  source=0