DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous Driving

Paper

DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous Driving

Demo

Getting Started

Environmental Setups

git clone https://github.com/EnVision-Research/DriveRecon.git --recursive
cd DriveRecon
conda env create -f environment.yml

pip install -e submodules/depth-diff-gaussian-rasterization
pip install -e submodules/simple-knn

Preparing Dataset

Follow detailed instructions in Prepare Dataset.

Training

Single machines

accelerate  launch --config_file ./acc.yaml \
train.py  --port 6017 --expname 'waymo' --configs 'arguments/nvs.py'

Multiple machines and multiple gps

accelerate launch --config_file ./acc_config.yaml \
--machine_rank $MLP_ROLE_INDEX --num_machines 3 --num_processes 24 \
--main_process_ip $MLP_WORKER_0_HOST --main_process_port $MLP_WORKER_0_PORT  \
train.py --port 6017 --expname 'waymo' \
--configs 'arguments/nvs.py'

Evaling

# python eval.py --checkpoint_path "./checkpoint_10000.pth" --port 6017 --expname 'waymo' --configs 'arguments/nvs.py'

Citation

If you find this project helpful, please consider citing the following paper:

@article{Lu2024DrivingRecon,
        title={DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous Driving},
        author={Hao LU, Tianshuo XU, Wenzhao ZHENG, Yunpeng ZHANG, Wei ZHAN, Dalong DU, Masayoshi Tomizuka, Kurt Keutzer, Yingcong CHEN},
        journal={arXiv preprint arXiv:2412.09043},
        year={2024}
      }