目录
2016 You Only Look Once: Unified, Real-Time Object Detection(CVPR,Joseph Redmon)
2017 YOLO9000: Better, Faster, Stronger (CVPR,Joseph Redmon)
2018 YOLOv3:AnIncrementalImprovemen (CVPR,Joseph Redmon)
2020 YOLOv4: Optimal Speed and Accuracy of Object Detection
2022 YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications (2022)
2023 YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
2024 YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information
2020 PP-YOLO: An Effective and Efficient Implementation of Object Detector
2021 PP-YOLOv2: A Practical Object Detector
2021 YOLOX: Exceeding YOLO Series in 2021
2021 YOLO-ReT: Towards High Accuracy Real-time Object Detection on Edge GPU
团队 | base | base |
Joseph Redmon | v1,v2,v3 | |
Alexey Bochkovskiy | v4,v7,v9 | |
Ultralytics | v5,v8,v11 | |
美团视觉智能部 | v6 | |
清华大学团队 | v10 | |
其他 | V12 | v11 |
2016 You Only Look Once: Unified, Real-Time Object Detection(CVPR,Joseph Redmon)
输入:448x448 45FPS 63.4mAP
输出:(4+1+4+1+20)*[7x7]
预测 [x,y,w,h,confidence, x1,y2,w2,h2,confidence2,classScore*20], 非anchor形式
2017 YOLO9000: Better, Faster, Stronger (CVPR,Joseph Redmon)
使用anchor的预测形式
输出:(4+1+20)*[5*(13*13)]
[tx,ty,tw,th,confidence, class*20]*5 ,每个框5个anchor
2018 YOLOv3:AnIncrementalImprovemen (CVPR,Joseph Redmon)
多尺度输出
输出:(4+1+80)*[3*(13*13+26*26+52*52)] ,每个框3个anchor
N*N*[3*(1+4+80)] ,每个网格预测三个anchor box
2020 YOLOv4: Optimal Speed and Accuracy of Object Detection
对V3进行了优化
CSPDarknet53 主干,Mish 激活,Dropblock正则化
SPP模块,FPN+PAN
输出:(4+1+80)*[3*(13*13+26*26+52*52)] ,每个框3个anchor
2021 YOLO V5 (Ultralytics)
输出:(4+1)*[3*(20*20+40*40+80*80)]
2022 YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications (2022)
2023 YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
2024 YOLO V8 (Ultralytics)
2024 YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information
YOLO V10
YOLO V11 (Ultralytics)
YOLO V12
输出 (4+80)x8400