Merge branch 'main' of github.com:WncFht/notes

Author: WncFht

docs/Technology/ffb6d.md

@@ -0,0 +1,220 @@
+# FFB6D环境配置指南：原生系统安装
+
+## 1. 系统要求
+
+- Ubuntu 20.04/22.04/24.04
+- NVIDIA GPU（支持CUDA）
+- 至少8GB内存
+- 至少30GB磁盘空间
+
+## 2. 基础环境配置
+
+### 2.1 安装NVIDIA驱动
+
+```bash
+# 添加NVIDIA包仓库
+sudo add-apt-repository ppa:graphics-drivers/ppa
+sudo apt-get update
+
+# 安装NVIDIA驱动
+sudo apt-get install -y nvidia-driver-535  # 根据需要选择版本
+
+# 重启系统
+sudo reboot
+
+# 验证安装
+nvidia-smi
+```
+
+### 2.2 安装CUDA和cuDNN
+
+```bash
+# 下载并安装CUDA 11.0
+wget https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux.run
+sudo sh cuda_11.0.3_450.51.06_linux.run
+
+# 配置环境变量
+echo 'export PATH=/usr/local/cuda-11.0/bin:$PATH' >> ~/.bashrc
+echo 'export LD_LIBRARY_PATH=/usr/local/cuda-11.0/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
+source ~/.bashrc
+
+# 下载并安装cuDNN 8.0
+# 注：需要从NVIDIA开发者网站下载cuDNN v8.0，解压后：
+sudo cp cuda/include/cudnn*.h /usr/local/cuda/include
+sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
+sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*
+```
+
+### 2.3 安装系统依赖
+
+```bash
+sudo apt-get update
+sudo apt-get install -y \
+    python3.6 \
+    python3.6-dev \
+    python3-pip \
+    git \
+    cmake \
+    build-essential \
+    libopencv-dev \
+    libglib2.0-0 \
+    libsm6 \
+    libxext6 \
+    libxrender-dev \
+    libboost-all-dev \
+    libeigen3-dev
+```
+
+### 2.4 配置Python环境
+
+```bash
+# 设置Python 3.6为默认版本
+sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.6 1
+sudo update-alternatives --set python3 /usr/bin/python3.6
+
+# 配置pip镜像源
+pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
+
+# 升级pip
+python3 -m pip install --upgrade pip
+```
+
+## 3. 安装PyTorch和依赖包
+
+### 3.1 安装PyTorch
+
+```bash
+pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 -f https://download.pytorch.org/whl/torch_stable.html
+```
+
+### 3.2 安装项目依赖
+
+创建requirements.txt并安装依赖：
+
+```bash
+pip3 install -r requirements.txt
+```
+
+requirements.txt内容：
+```
+h5py         
+numpy              
+pyyaml==5.4.1
+enum34        
+future           
+scipy==1.4.1          
+opencv_contrib_python==3.4.2.16               
+transforms3d==0.3.1     
+scikit_image==0.13.1     
+lmdb==0.94              
+setuptools==41.0.0    
+cffi==1.11.5         
+easydict==1.7      
+plyfile==0.6      
+pillow==8.2.0 
+dataclasses
+glumpy                                                      
+tqdm
+tensorboardX 
+pandas
+scikit-learn         
+scipy 
+termcolor
+pybind11
+```
+
+## 4. 编译和安装特殊组件
+
+### 4.1 编译apex
+
+```bash
+git clone https://github.com/NVIDIA/apex
+cd apex
+export TORCH_CUDA_ARCH_LIST="6.0;6.1;6.2;7.0;7.5"
+python setup.py install -v
+cd ..
+```
+
+### 4.2 安装和编译normalspeed
+
+```bash
+# 1. 准备OpenCV环境
+pip uninstall opencv-python opencv-python-headless -y
+pip install opencv-python==4.5.3.56
+
+# 2. 克隆并安装normalspeed
+git clone https://github.com/hfutcgncas/normalspeed.git
+cd normalspeed/normalSpeed
+
+# 安装编译依赖
+sudo apt-get install python3-pybind11
+pip3 install Cython==0.29.15
+
+# 清理并重新安装
+rm -rf build/ dist/ *.egg-info/
+python3 setup.py install --user
+cd ../..
+```
+
+## 5. 克隆和配置FFB6D
+
+```bash
+# 克隆代码
+git clone https://github.com/ethnhe/FFB6D.git
+cd FFB6D
+
+# 创建必要的目录
+mkdir -p datasets models train_log
+
+# 配置环境变量
+export PYTHONPATH=$PYTHONPATH:$(pwd)
+```
+
+## 6. 验证安装
+
+```bash
+# 验证CUDA支持
+python3 -c "import torch; print('CUDA available:', torch.cuda.is_available())"
+
+# 验证apex安装
+python3 -c "from apex import amp; print('APEX installed')"
+
+# 验证normalspeed安装
+python3 -c "import normalSpeed; print('normalspeed installed')"
+```
+
+## 7. 常见问题
+
+### 7.1 网络问题
+
+```bash
+# 使用代理（如需要）
+export http_proxy="http://proxy:port"
+export https_proxy="http://proxy:port"
+
+# 或使用国内镜像源
+pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
+```
+
+### 7.2 版本兼容性问题
+
+- 确保NVIDIA驱动版本支持CUDA 11.0
+- 确保Python包版本相互兼容
+- 检查CUDA版本与PyTorch版本的匹配
+
+### 7.3 编译错误
+
+- 确保已安装所有必要的编译工具
+- 检查CUDA路径配置是否正确
+- 确认系统库版本是否满足要求
+
+## 8. 训练
+
+按照官方文档配置LineMOD数据集并开始训练。
+
+## 参考资料
+
+1. [FFB6D项目](https://github.com/ethnhe/FFB6D)
+2. [CUDA安装指南](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html)
+3. [PyTorch安装指南](https://pytorch.org/get-started/locally/)
+4. [【论文笔记】FFB6D | 马浩飞丨博客](https://www.mahaofei.com/post/d027527.html)