1. 安装 Cuda 11.8
sudo apt update
sudo apt install build-essential
安装cuda,这里的wget也可以在简介中的网盘下载
wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
sudo sh cuda_11.8.0_520.61.05_linux.run
安装完成后,通过下面指令打开vscode:
code .
然后再文件最后加入:
export PATH=/usr/local/cuda-11.8/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:$LD_LIBRARY_PATH
2. 安装colmap
为了支持调用CUDA的colmap,我们需要自己进行编译,首先安装各种依赖
sudo apt-get update && sudo apt-get install -y git cmake build-essential ninja-build libboost-program-options-dev libboost-filesystem-dev libboost-graph-dev libboost-system-dev libboost-test-dev libeigen3-dev libfreeimage-dev libgoogle-glog-dev libgflags-dev libglew-dev libqt5opengl5-dev qt5-default libatlas-base-dev libsuitesparse-dev libcgal-dev libcgal-qt5-dev libceres-dev libflann-dev liblz4-dev libsqlite3-dev libmetis-dev
然后用gitclone拉取代码,并切换分支:(也可以从百度云下载)
git clone https://github.com/colmap/colmap.git
cd colmap # 强烈建议切换到 3.9.1 分支,比 dev 分支更稳定,不容易报错
git checkout 3.9.1
编译:
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
ninja # 这里会跑很久,但是有一个进度条,等待进度条走完就行了
sudo ninja install
3. 测试colmap(可选)
使用我提供的数据进行测试,也可以自己拍
cd ~/code/script
sudo chmod +x mapping.sh
./mapping.sh ~/code/data/
4. Miniconda
下载 Miniconda 安装脚本,或者直接从百度云下载
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc
4. 安装pytorch
查看当前conda的源
conda config --show-sources
删除掉系统的配置文件:
rm /home/mz/miniconda3/.condarc
输入:
code ~/.condarc
编辑为:
channels:
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
- nodefaults
show_channel_urls: true
创建一个虚拟环境
conda create -n gsplat python=3.10
conda activate gsplat
将 pip 源永久设置为清华源
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
安装pytorch
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
验证是否成功:
python
> import torch
> torch.cuda.is_available()
5. 安装gsplat
pip install gsplat
为了训练,我还需要将gsplat的源码下载下来:
git clone https://github.com/nerfstudio-project/gsplat.git
cd gsplat/examples
安装其他训练所需要的库:
pip install -r requirements.txt --no-build-isolation
开始训练:
python simple_trainer.py default --data_dir ~/code/script/3dgs_project/output/ --data_factor 1 --save_ply
