.
├── README.md
├── LICENSE.md
├── assets
├── environment.* # conda venv
├── SIBR_viewers_bin # SIBR viewer
├── submodules # dependent git repos (e.g. dgr engines)
│ ├── diff-gaussian-rasterization
│ ├── diff-gaussian-rasterization-kahsolt
│ ├── simple-knn
│ └── init_repos.cmd
├── data # dataset inputs
├── output # result ouputs
├── modules
│ ├── data # dataloaders
│ ├── lpipsPyTorch # LPIPS metric
│ ├── hparam.py # default hparams
│ ├── camera.py # default camera
│ ├── scene.py # default scene
│ ├── layers.py # common neural network layers/blocks/modules
│ ├── utils # common utils
│ ├── network_gui.py # remote monitor for SIBR viewer
│ └── morphs # implementing 3d-gs variations
│ ├── README.md
│ └── <morph>
│ ├── model.py
│ ├── render.py
│ ├── train.py
│ ├── hparams.py # override default
│ ├── camera.py # override default
│ ├── scene.py # override default
│ └── ...
│── run.cmd # example run commands
├── train.py # train entry script
├── render.py # render entry script
├── metrics.py # metric entry script
├── convert.py # dataset convert entry script (not used)
└── full_eval.py # full eval entry script (not used)
⚪ 引用关系
- Scene: 场景(顶级对象,一个实验,一切信息都能通过它取得)
- HyperParams: 超参
- Camera: 相机真值
- GaussianModel: 可训练的高斯模型
⚪ 构造关系
hp = HyperParams()
scene = Scene(hp)
scene.train_cameras: List[Camera]
scene.test_cameras: List[Camera]
scene.gaussians: GaussianModel