[CVPR 2026] VAD-GS: Visibility-Aware Densification for 3D Gaussian Splatting in Dynamic Urban Scenes

Project page | Paper | Youtube | Bilibili

Installation

Clone this repository and checkout dev branch

git clone https://github.com/YikangZhang1641/VAD-GS.git
git checkout -b dev origin/dev

Build tools

1. Install COLMAP (tested version 3.10-dev)
2. Build SIBR_viewer following the tutorial of 3DGS

Set up the environment

# Set conda environment
conda create -n vadgs python=3.8
conda activate vadgs

# Install torch (corresponding to your CUDA version)
pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113

# Install requirements
pip install -r requirements.txt

# Install submodules
pip install ./submodules/diff-gaussian-rasterization
pip install ./submodules/simple-knn
pip install ./submodules/simple-waymo-open-dataset-reader
pip install ./submodules/MyPropagation

Datasets

data/
   ├── nuscenes/
   │     ├── raw/
   │     ├── processed_10Hz/
   │     │     ├── mini/
   │     │     │     ├── 000/
   │     │     │     │     ├── images/
   │     |     │     │     ├── ego_pose/
   │     |     │     │     ├── lidar_depth/
   │     |     │     │     └── ...
   │     │     │     ├── 001/
   │     │     │     ├── ...
   └── waymo/
         |...

Example

• We provide a nuScenes example here. Download and extract it to the folder path above.

For training:

python train.py --config configs/example/nuscenes_train_000.yaml

To generate visual outputs:

python render.py --config configs/example/nuscenes_train_000.yaml mode evaluate

For evaluation:

python metrics.py --config configs/example/nuscenes_train_000.yaml