基本和第一个一样的流程
cuda环境安装教程
https://www.cnblogs.com/gooutlook/p/17677113.html
工程环境 安装指令
# 官网 https://github.com/graphdeco-inria/reduced-3dgs # ============= 1从文件创建环境 ============ 容易在submodules安装时候报错卡死 conda env create --file environment.yml # 删除环境指令(如果需要) conda env remove --name gaussian_splatting # ============= 2手动创建 =================== # 2-1 创建环境 conda create --name gaussian_splatting # 2-2 激活环境 conda activate gaussian_splatting # 2-3 安装在线库 pip install --upgrade setuptools wheel pip install --upgrade pip==22.3.1 pip install plyfile pip install tqdm pip install urllib3==2.2.1 pip install pandas pip install torch==1.12.1+cu116 torchaudio==0.12.1+cu116 torchvision==0.13.1+cu116 -f https://download.pytorch.org/whl/torch_stable.html # 2-4 安装离线库 代码文件夹自带的库 #pip install ./submodules/diff-gaussian-rasterization 无效 #pip install ./submodules/simple-knn 无效 # 需要用sudo模式, 同时指定目标conda环境的python去安装 sudo /home/dongdong/1sorftware/1work/yes/envs/gaussian_splatting/bin/python3.7 -m pip install submodules/simple-knn sudo /home/dongdong/1sorftware/1work/yes/envs/gaussian_splatting/bin/python3.7 -m pip install submodules/diff-gaussian-rasterization # 0查看内存情况====================== watch -n 1 nvidia-smi # 1激活环境========================== conda activate py37gaosi 老版本 conda activate gaussian_splatting 最新版本 # 2训练============================= 训练时候 其他占用显卡内存的软件需要关掉例如colmap 不然内存会爆满 #--resolution 2 原来图像的1/2分辨率 如果超过1600分辨率, 默认强行缩放,除非指定参数为1才是原分辨率 #-s 数据文件夹 #-m 训练结果 #--data_device cpu gpu 选择训练模式 #--sh_degree 球谐函数 默认情况下,球谐函数具有学习率0.0025。 给成0 减少内存 python ./train.py -s ../data/tandt/truck/ -m ../data/tandt/truck/train_out/ python ./train.py -s ./data/gaosi_0559/ -m ./data/gaosi_0559/train_out/ --resolution 2 --data_device cpu --sh_degree 0 python ./train.py -s /home/dongdong/2project/0data/ring/sence2/ -m /home/dongdong/2project/0data/ring/sence2/train_out/ --data_device cpu # 3可视化============================== # 3-1 训练过程中查看 sudo ./SIBR_viewers/install/bin/SIBR_remoteGaussian_app # 3-2 训练以后查看 sudo ./SIBR_viewers/install/bin/SIBR_gaussianViewer_app -m ./data/tandt/truck/train_out sudo ./SIBR_viewers/install/bin/SIBR_gaussianViewer_app -m /home/dongdong/2project/0data/ring/sence3/train_out/ sudo ./SIBR_viewers/install/bin/SIBR_gaussianViewer_app -m /home/dongdong/2project/0data/ring/sence4/train_out/
编译可视化软件
cd SIBR_viewers cmake -Bbuild . cmake --build build --target install --config RelWithDebInfo
如果opencv不是安装在默认系统环境,手动指定
set(CMAKE_PREFIX_PATH "/home/dongdong/1sorftware/1work/opencv/opencv455/install")
opencv 安装路径
标签:SIBR,高斯,--,gaussian,train,install,pip,版本,3D From: https://www.cnblogs.com/gooutlook/p/18377816