1.环境准备

1.1.下载diff-gaussian-rasterization

这里本来没啥说的，直接从github上下载就行了，但是我踩坑了，下的版本不对，后续运行报错参数个数对不上，特在此给大家避坑，注意一定要下带3dgs版本的diff-gaussian-rasterization，否则后续会报错。

或者直接用我贴的下载地址：

https://github.com/graphdeco-inria/diff-gaussian-rasterization/tree/3dgs_accel

不要直接下载main版本的!不要直接下载main版本的!不要直接下载main版本的!

否则就会跟我一样报错：

Number of points at initialisation :  107 [06/08 17:53:37]
Training progress:   0%|                                                                                                                                                                | 0/30000 [00:00<?, ?it/s]Traceback (most recent call last):File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/train.py", line 282, in <module>training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from)File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/train.py", line 111, in trainingrender_pkg = render(viewpoint_cam, gaussians, pipe, bg, use_trained_exp=dataset.train_test_exp, separate_sh=SPARSE_ADAM_AVAILABLE)File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/gaussian_renderer/__init__.py", line 36, in renderraster_settings = GaussianRasterizationSettings(
TypeError: GaussianRasterizationSettings.__new__() got an unexpected keyword argument 'antialiasing'

Traceback (most recent call last):File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/train.py", line 282, in <module>training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from)File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/train.py", line 111, in trainingrender_pkg = render(viewpoint_cam, gaussians, pipe, bg, use_trained_exp=dataset.train_test_exp, separate_sh=SPARSE_ADAM_AVAILABLE)File "/mnt/geovis/zhouy/001_project/gaussian-splatting-main/gaussian_renderer/__init__.py", line 102, in renderrendered_image, radii, depth_image = rasterizer(
ValueError: not enough values to unpack (expected 3, got 2)
Training progress:   0%|

1.2.下载glm

这个没啥特殊注意，直接在github上下载就行

命令下载：

git clone https://github.com/icaven/glm.git

下载完成后将其放在./gaussian-splatting-main/submodules/diff-gaussian-rasterization/third_party/目录下，文件夹名称为glm，如下图

1.3.安装diff-gaussian-rasterization

cd 到diff-gaussian-rasterization目录下，运行：

python setup.py install

正常安装的话应该会如下图所示，有一些warning，不用管它。直至出现Installed /root/anaconda3/envs/gaussian/lib/python3.10/site-packages/diff_gaussian_rasterization-0.0.0-py3.10-linux-x86_64.egg
Processing dependencies for diff-gaussian-rasterization==0.0.0
Finished processing dependencies for diff-gaussian-rasterization==0.0.0

1.4.安装simple-knn

安装没啥注意的，建议都下载下来离线包，再上传上服务器解压安装。

下载地址：

https://github.com/camenduru/simple-knn

将下载的压缩包上传至./gaussian-splatting-main/submodules/下，解压，运行以下命令即可完成安装：

git clone https://github.com/camenduru/simple-knn && cd simple-knn
pip install .

1.5.安装fused-ssim

同上，先下载离线文件，上传服务器，解压

下载地址：

https://github.com/rahul-goel/fused-ssim

cd到./gaussian-splatting-main/submodules/fused-ssim目录下，运行以下代码实现安装：

pip install . --no-build-isolation

2.训练过程

(gaussian) [root@t97 gaussian-splatting-main]# python train.py -s ./data -m ./output/mydata
Optimizing ./output/mydata
Output folder: ./output/mydata [06/08 18:21:32]
Tensorboard not available: not logging progress [06/08 18:21:32]
Reading camera 4/4 [06/08 18:21:32]
Loading Training Cameras [06/08 18:21:32]
Loading Test Cameras [06/08 18:21:32]
Number of points at initialisation :  107 [06/08 18:21:32]
Training progress:  23%|??????????????????????????                                                                                    | 6990/30000 [00:34<02:03, 186.02it/s, Loss=0.0057908, Depth Loss=0.0000000]
[ITER 7000] Evaluating train: L1 0.004467152804136276 PSNR 40.17698516845704 [06/08 18:22:07][ITER 7000] Saving Gaussians [06/08 18:22:07]
Training progress: 100%|?????????????????????????????????????????????????????????????????????????????????????????????????????????????| 30000/30000 [02:59<00:00, 167.57it/s, Loss=0.0038100, Depth Loss=0.0000000][ITER 30000] Evaluating train: L1 0.002960224240086973 PSNR 44.79262390136719 [06/08 18:24:31][ITER 30000] Saving Gaussians [06/08 18:24:31]Training complete. [06/08 18:24:33]