YOLOv5结构讲解

爱吃花生米的枫叶

已于 2022-05-24 13:50:10 修改

阅读量4.5k

点赞数 3

文章标签：深度学习

于 2021-12-07 20:42:31 首次发布

本文链接：https://blog.csdn.net/weixin_42035080/article/details/121777772

版权

YOLOv5关闭wandb

首先打开utils/loggers/wandb/wandb_utils.py

在32行新添加wandb = None

然后打开utils/loggers/__init__.py

在25行新添加wandb = None

YOLOv5-6.0中各模型区别

yolov5n, yolov5s, yolov5m, yolov5l, yolov5x的模型只在模型深度（C3模块数）和宽度（网络的通道数）上不同，其他都一样。

以yolov5l为基准，其他模型以此放缩。

进入目录model/ ，命令行输入：grep "multiple" *.yaml

yolov5l.yaml:depth_multiple: 1.0  # model depth multiple
yolov5l.yaml:width_multiple: 1.0  # layer channel multiple
yolov5m.yaml:depth_multiple: 0.67  # model depth multiple
yolov5m.yaml:width_multiple: 0.75  # layer channel multiple
yolov5n.yaml:depth_multiple: 0.33  # model depth multiple
yolov5n.yaml:width_multiple: 0.25  # layer channel multiple
yolov5s.yaml:depth_multiple: 0.33  # model depth multiple
yolov5s.yaml:width_multiple: 0.50  # layer channel multiple
yolov5x.yaml:depth_multiple: 1.33  # model depth multiple
yolov5x.yaml:width_multiple: 1.25  # layer channel multiple

	depth	width
yolov5n	0.33	0.25
yolov5s	0.33	0.5
yolov5m	0.67	0.75
yolov5l	1	1
yolov5x	1.33	1.25

YOLOv5-6.0对比YOLOv5-5.0

https://github.com/ultralytics/yolov5/releases/tag/v6.0https://github.com/ultralytics/yolov5/releases/tag/v6.0

Model Architecture: Updated backbones are slightly smaller, faster and more accurate.

Replacement of Focus() with an equivalent Conv(k=6, s=2, p=2) layer (#4825 by @thomasbi1) for improved exportability
New SPPF() replacement for SPP() layer for reduced ops (#4420 by @glenn-jocher)
Reduction in P3 backbone layer C3() repeats from 9 to 6 for improved speeds
Reorder places SPPF() at end of backbone
Reintroduction of shortcut in the last C3() backbone layer
Updated hyperparameters with increased mixup and copy-paste augmentation

在backbone有更改。

SPPF对比SPP

位于文件：models/common.py

两者结果是一样，SPPF(Fast)在于：重用了上一个maxpool的结果，第一次maxpool5，第二次在此基础上继续maxpool5则等同于maxpool9，第三次在此基础上继续maxpool5则等同于maxpool13，最后拼接原x，maxpool5结果，maxpool9结果，maxpool13结果，再用一个卷积调制通道与输入相同。