CRUISE: Cooperative Reconstruction and Editing in V2X Scenarios using Gaussian Splatting

---

Demo

demo.mp4

Environment

Configure Python environment

# conda environment
conda create -n V2Xsim python=3.8
conda activate V2Xsim

# CUDA 11.8
pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cu118

# Install requirements
pip install -r requirments.txt

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

Configure environment for generating masks GroudingDINO, and download the SAM checkpoint

Configure Python environment of TRELLIS

Dataset

• Download the original dataset: DAIR-V2X-SPD

  • Use data_process.ipynb for data pre-processing

  • Use generate_mask.ipynb to generate sky mask and ego mask

• Or you can directly download and use the processed synthetic dataset: (comming soon)

• Download the high-quality vehicle dataset for Relightable3DGaussian: (comming soon)

Training

If you want to modify the training command, change the content in train.sh and specify the corresponding config.

./script/train.sh

Rendering

Use following command to render.

python render.py --config configs/xxxx.yaml mode edit

Generate Synthetic dataset

Use the following command to perform preliminary organization and packaging of the render data.

python generate_dataset.py

Then use the command below to merge the synthetic dataset with the original dataset for downstream tasks.

python append_dataset.py

Downstream tasks

Please complete the corresponding downstream as shown in the corresponding document.

• Infrastructure ciew 3d object detection: BEVHeight: https://github.com/ADLab-AutoDrive/BEVHeight

• Vehicle ciew 3d object detection:Monolss: https://github.com/Traffic-X/MonoLSS

• Coolaborative view 3d object detection: https://github.com/AIR-THU/DAIR-V2X/tree/main/configs/vic3d-spd/late-fusion-image