YOLO-Master运行容器配置方法

拉取基础镜像

docker pull swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/nvidia/cuda:12.6.0-devel-ubuntu22.04
docker tag swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/nvidia/cuda:12.6.0-devel-ubuntu22.04 docker.io/nvidia/cuda:12.6.0-devel-ubuntu22.04
docker run --shm-size=“32g” --gpus all -it -v /media:/media -v /mnt:/mnt --name yolo_test0630 nvidia/cuda:12.6.0-devel-ubuntu22.04 /bin/bash

以下在容器中运行

安装 Miniforge3

./Miniforge3-25.3.0-3-Linux-x86_64.sh
source ~/miniforge3/bin/activate
source ~/.bashrc
conda init bash

创建虚拟环境

conda create -n yolo_master python=3.11 -y
conda activate yolo_master

查看 CUDA 版本

nvcc -V

pip install torch2.6.0 torchvision0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu126
cd YOLO-Master-main/
pip install -r requirements.txt
pip install -e .

安装 flash-attention 加速库对 CUDA 版本的要求较高

pip install packaging
pip install ninja
pip install opencv-python-headless==4.13.0.90
MAX_JOBS=4 pip install flash-attn --no-build-isolation --no-cache-dir

问题：ImportError: libxcb.so.1: cannot open shared object file: No such file or directory

ImportError: libxcb.so.1 是 Linux 无图形界面环境（服务器/容器）缺少 OpenCV 依赖的 X11 图形库，cv2 默认绑定 GUI 模块，找不到 libxcb 动态库。
解决：conda install -c conda-forge libxcb