3D gaussian-splatting代码在ubuntu20.04 & CUDA11.8下的配置

修改

241008：增加了ubuntu环境下的可视化SIBR_viewers配置3D gaussian-splatting可视化SIBR_viewers在ubuntu20.04下的配置-CSDN博客

本文参考&致谢：

【3DGS】Ubuntu20.04系统搭建3D Gaussian Splatting及可视化环境_ubuntu 3dgs安装-CSDN博客

Ubuntu20.04 3DGS复现全流程_ubuntu20.04安装3dgs-CSDN博客

代码来源：

GitHub - graphdeco-inria/gaussian-splatting: Original reference implementation of "3D Gaussian Splatting for Real-Time Radiance Field Rendering"

基础环境：

1. ubuntu 20.04，架构x86_64

hostnamectl

(base) root@test:~$ hostnamectl
Icon name: computer-desktop
Chassis: desktop
Operating System: Ubuntu 20.04.6 LTS
Kernel: Linux 5.15.0-117-generic
Architecture: x86-64

2. g++和gcc，输入g++ -v和gcc -v

g++ -v
gcc -v

(base) root@test:~$ g++ -v

Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/9/lto-wrapper
OFFLOAD_TARGET_NAMES=nvptx-none:hsa
OFFLOAD_TARGET_DEFAULT=1
Target: x86_64-linux-gnu
Configured with: ......
Thread model: posix
gcc version 9.4.0 (Ubuntu 9.4.0-1ubuntu1~20.04.2)

(base) root@test:~$ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/9/lto-wrapper
OFFLOAD_TARGET_NAMES=nvptx-none:hsa
OFFLOAD_TARGET_DEFAULT=1
Target: x86_64-linux-gnu
Configured with: ..........
Thread model: posix
gcc version 9.4.0 (Ubuntu 9.4.0-1ubuntu1~20.04.2)

3. 显卡驱动版本535.183.01，

nvidia-smi

4. CUDA版本11.8

nvcc -V

(base) root@test:~$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

CUDA如果已下载，跳过以下的下载过程：

(1) CUDA11.8版本下载

下载地址https://developer.nvidia.com/cuda-11-8-0-download-archive

(2) 输入图中的命令sudo sh cuda_11.8.0_520.61.05_linux.run后，等待一会，出现以下命令框，选择continue

(3) 输入accept

(4) 如果已安装显卡驱动（见上）且驱动版本满足CUDA11.8的要求，要取消勾选Driver

(5) 等待安装。安装成功后，输入以下命令：

ls /usr/local/

(base) root@test:~$ ls /usr/local/
bin  cuda  cuda-11.8  etc  games  include  lib  man  sbin  share  src  sunlogin

(6) 可以看到已经成功安装cuda-11.8，但还需要配置环境变量

vim ~/.bashrc

(base) root@test:~$ vim ~/.bashrc

(7) 添加以下环境变量内容（重要）

export PATH=/usr/local/cuda-11.8/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

 输入vim指令 :wq 保存更改

(8) 输入以下指令，使上述环境变量生效

source ~/.bashrc

 (base) root@test:~$ source ~/.bashrc

(9) 添加完后，输入nvcc -V，可看到cuda版本信息

nvcc -V

(base) root@test:~$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

(10) 输入which nvcc，可看到nvcc安装位置

which nvcc

(base) root@test:~$ which nvcc
/usr/local/cuda-11.8/bin/nvcc

至此，CUDA安装成功

配置3D Gaussian Splatting环境

1. 拉取github代码，注意带上--recursive后缀（会拉取子模块的代码）。

git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive

2. 拉取后查看以下两个文件夹中 是否是非空

gaussian-splatting/submodules/diff-gaussian-rasterization

gaussian-splatting/submodules/diff-gaussian-rasterization/third_party/glm

3. 查看environment.yaml文件 原内容

name: gaussian_splatting
channels:
  - pytorch
  - conda-forge
  - defaults
dependencies:
  - cudatoolkit=11.6
  - plyfile
  - python=3.7.13
  - pip=22.3.1
  - pytorch=1.12.1
  - torchaudio=0.12.1
  - torchvision=0.13.1
  - tqdm
  - pip:
    - submodules/diff-gaussian-rasterization
    - submodules/simple-knn

4. 本文不采用以上配置，因为在执行pip submodules时，可能会卡住。

(1) 分别执行命令：

cd gaussian-splatting

conda create -n gaussian_splatting python=3.8

conda activate gaussian_splatting

(2) 安装torch，Previous PyTorch Versions | PyTorch

pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118

经过漫长地等待，输入以下命令验证torch 和CUDA可用

python

import torch

print(torch.cuda.is_available())

exit()

(gaussian_splatting) root@test:~/postg/thesis/gaussian-splatting$ python
Python 3.7.13 (default, Oct 18 2022, 18:57:03) 
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True

>>> exit()

(3) 执行安装子模块，这一步容易报错

pip install submodules/diff-gaussian-rasterization

pip install submodules/simple-knn

可能会报错，报错信息： 

cc1plus: fatal error: cuda_runtime.h: No such file or directory

gcc: fatal error: cannot execute ‘cc1plus’: execvp: No such file or directory

packages/torch/utils/cpp_extension.py", line 1785, in _write_ninja_file_and_compile_objects _run_ninja_build( File "/home/gyw/anaconda3/envs/pgsr/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 2121, in _run_ninja_build raise RuntimeError(message) from e RuntimeError: Error compiling objects for extension [end of output] note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed building wheel for diff_plane_rasterization Running setup.py clean for diff_plane_rasterization Failed to build diff_plane_rasterization ERROR: ERROR: Failed to build installable wheels for some pyproject.toml based projects (diff_plane_rasterization)

先查看是否已安装c语言编译环境（见上）

gcc -v
g++ -v

(gaussian_splatting) root@test:~/postg/thesis/gaussian-splatting$ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/9/lto-wrapper
OFFLOAD_TARGET_NAMES=nvptx-none:hsa
OFFLOAD_TARGET_DEFAULT=1
Target: x86_64-linux-gnu
Configured with:
Thread model: posix
gcc version 9.4.0 (Ubuntu 9.4.0-1ubuntu1~20.04.2) 

注意红色的部分，如果红色部分是某个conda环境的路径，就删除此conda环境

如果已安装，再查看是否已配置bashrc中的环境变量（见上）

export PATH=/usr/local/cuda-11.8/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

(4) 执行安装剩余包

pip install plyfile tqdm

下载数据集并训练

参考 下面链接中的【下载数据集并训练】【3DGS】Ubuntu20.04系统搭建3D Gaussian Splatting及可视化环境_ubuntu 3dgs安装-CSDN博客

下载后的数据集 tandt_db.zip 如下，并新建data文件夹，将 tandt_db.zip 其中的truck文件复制到data文件夹中

python train.py -s data/truck/ -m data/output

会自动生成输出结果文件夹，结果的树结构如下

(base) root@test:~/postg/thesis/gaussian-splatting/data$ tree output
output
    ├── cameras.json
    ├── cfg_args
    ├── input.ply
    └── point_cloud
        ├── iteration_30000
        │   └── point_cloud.ply
        └── iteration_7000
            └── point_cloud.ply

4 directories, 5 files

可视化结果

gaussian-splatting文件夹中SIBR_viewers是用来可视化结果的应用，可用于ubuntu下的可视化

ubuntu下的可视化配置麻烦（以后实现），本文直接在Windows下进行可视化，将 结果文件 output 打包为zip并复制到Windows系统下（可通过mobaxterm软件进行传递，且无需打包）。注意，Windows下的显卡算力要大于7.0，https://developer.nvidia.com/cuda-gpus 查看

Windows下的可视化SIBR_viewers应用下载https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/binaries/viewers.zip

将output文件移动至viewers下

cmd命令

.\bin\SIBR_gaussianViewer_app.exe -m .\output

等待，即可视化成功