DINOv3权重文件下载  https://blog.csdn.net/qq_53644346/article/details/151717501  

https://blog.csdn.net/qq_43619271/article/details/150775085

GitHub  https://github.com/facebookresearch/dinov3/tree/main

两个示例：

https://github.com/yifangao112/DinoUNet.git

https://github.com/WZH0120/DINOv3-UNet

https://github.com/WZH0120/SAM2-UNet

https://github.com/WZH0120/SAM3-UNet

https://github.com/sayedmohamedscu/DinoV3-visualization-applications

____________________________________________________________________

在Featurize算力平台跑的配置：

pip list
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
#安装完毕查看
pip list
conda env list

#3.安装必要的依赖
pip install imageio
pip install opencv-python matplotlib pandas tqdm
#执行过程包错：ModuleNotFoundError: No module named 'torchmetrics'
pip install torchmetrics -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install termcolor colorama
-------------------------------------
############## 跑dinov3代码 ################################
#1.下载dinov3源码
git clone --recurse-submodules https://github.com/facebookresearch/dinov3.git
#2.下载预训练模型：
featurize dataset download 20b02a11-83e8-4171-ab55-1519cd8d90e9
解压到根目录dinov3_vitl16_pretrain_lvd1689m-8aa4cbdd.pth，解压路径要相应修改DINOv3-UNet.py
#3.准备数据集：featurize dataset download 9f6a28d9-643e-4658-a81b-6f8e5deec7d4
解压dataset.zip到根目录，文件名就是dataset
#4.尝试运行训练代码：
python DINOv3_UNet_train.py
python DINOv3_UNet_test.py
--------------------------------------