用YOLO26训练自己的数据

刚睡醒YOLO26就悄悄发布了，看官方公布的折线图，MAP涨势凶猛，准备自己实测一下效果如何

1 环境配置

源码下载地址：YOLOv26

git clone --branch v8.4.0 https://github.com/ultralytics/ultralytics.git
cd ultralytics
conda create -n yolo26 python==3.9
conda activate yolo26
pip install -e . -i https://mirrors.huaweicloud.com/repository/pypi/simple/

等待依赖完成下载

2 数据集

这里准备了一个person的检测数据，简单划分后分布如下

data.yaml:

train:
- /home/workspace/dataset/Person/Images/Train

val:
- /home/workspace/dataset/Person/Images/Validate

test:
- /home/workspace/dataset/Person/Images/Test

is_coco: false
nc: 1
names:
- person

预训练模型：yolov

3 训练

yolo detect train model=yolo26n.pt data=/home/workspace/dataset/Person/data.yaml epochs=100 imgsz=640 device=0 batch=32