maple0leaves / SA-YOLOV7

基于Shuffle attention相似目标区分检测算法研究

SA-YOLOv7

paper on the way~

Performance

Dataset：hand_glove

Model	Size	P	R	map0.5:0.95	FPS
SA-YOLOv7	640	91.40%	82.50%	48.60%↑	72.2
YOLOv7	640	83.70%	82.30%	46.80%	89.2
YOLOv5m	640	89.30%	84.2%	47.8%	67.5
YOLOv5l	640	86.9%	88.00%	47.00%	59
YOLOv6n	640	79.10%	79.00%	46.30%	75
YOLOv6m	640	93.40%	70.00%	47.80%	46.6

Testing&Training

sa_yolov7.pt

I use pycharm connect to my server ,when I fix code ,I run it directly in my pycharm,do not use command in Terminal.If you like run in Terminal,mybe you should write commands by youself .

Testing

If you just want to try ,you should download sa_yolov7.pt,and change data/glove.yaml,test.py then run .if you want to test youself model ,you know ,just change paths .Also,you could ask me at issues.Welcome!

Training

Environment

pytorch2.0,cuda11.8,ubuntu20.04

RTX3080Ti *8

Reference

WongKinYiu/yolov7: Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors (github.com)

wofmanaf/SA-Net: Code for our ICASSP 2021 paper: SA-Net: Shuffle Attention for Deep Convolutional Neural Networks (github.com)

Last

If you have any questions ,you could ask me at issuces.Welcome! I very happy I could help you !

All of work finish by myself ,so if this work could help you ,please with my project link(https://github.com/maple0leaves/SA-YOLOV7/) Thank you!