Ultralytics yolo下载

1. 官方Github仓库

主仓库：https://github.com/ultralytics/ultralytics
模型权重托管仓库：https://github.com/ultralytics/assets/releases

2. 全部.pt预训练权重直链（v8.3.0版本，YOLOv8+YOLO11）

# yolov8l 检测 https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8l.pt # yolov8l-seg 实例分割 https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8l-seg.pt # yolo11n-cls 图像分类 https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n-cls.pt # yolo11n-pose 姿态估计 https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n-pose.pt # yolov8m-obb 旋转框检测 https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8m-obb.pt

3. 一键导出ONNX完整代码（复制运行）

第一步安装依赖

pipinstallultralytics

第二步导出脚本

fromultralyticsimportYOLO# 逐个加载pt并导出onnxmodel_list=["yolov8l.pt","yolov8l-seg.pt","yolo11n-cls.pt","yolo11n-pose.pt","yolov8m-obb.pt"]fornameinmodel_list:model=YOLO(name)# format=onnx 导出，默认640分辨率out_path=model.export(format="onnx")print(f"导出完成:{out_path}")

运行后，当前目录就会生成你需要的5个onnx文件：
yolov8l.onnx、yolov8l-seg.onnx、yolo11n-cls.onnx、yolo11n-pose.onnx、yolov8m-obb.onnx

4. 服务器上一键下载+导出全套（直接粘贴到Linux服务器执行）

# 下载权重wgethttps://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8l.ptwgethttps://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8l-seg.ptwgethttps://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n-cls.ptwgethttps://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n-pose.ptwgethttps://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8m-obb.pt# 安装依赖pipinstallultralytics# python导出onnxpython-c" from ultralytics import YOLO models = ['yolov8l.pt','yolov8l-seg.pt','yolo11n-cls.pt','yolo11n-pose.pt','yolov8m-obb.pt'] for m in models: p = YOLO(m).export(format='onnx') print('Exported:', p) "

5. 国内下载慢解决方案

如果github下载超时，可以使用加速地址：
格式：在原链接前面加https://mirror.ghproxy.com/
示例：

https://mirror.ghproxy.com/https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8l.pt

补充说明

1. 不要网上随便找别人打包好的onnx，输入尺寸、后处理节点各不相同，容易推理报错；官方pt导出最标准
2. 如果你需要固定onnx参数（例如简化op、固定输入shape、启用dynamic），我可以给你加上对应导出参数