This is a jittor implementation of DrivingForward.
git clone https://github.com/gotyao/DrivingForward_jittor
cd DrivingForward_jittor
conda create -n DFjittor python=3.9
conda activate DFjittor
pip install jittor
pip install -r requirements.txt
cd models/gaussian/gaussian_splatting_jittor/gaussian_renderer/diff_gaussian_rasterizater
cmake .
make -j
cd ../../scene/simple_knn
cmake .
make -j
cd ../../../..
Note: The repository uses Jittor_Perceptual-Similarity-Metric for evaluation. Please download the pretrained model from the source repository and place them within the DFjittor folder.
- Download nuScenes official dataset
- Place the dataset in
input_data/nuscenes/
Data should be as follows:
├── input_data
│ ├── nuscenes
│ │ ├── maps
│ │ ├── samples
│ │ ├── sweeps
│ │ ├── v1.0-test
│ │ ├── v1.0-trainval
Get the pretrained models, save them to the root directory of the project, and unzip them.
For SF mode, run the following:
python -W ignore eval.py --weight_path ./weights_SF --novel_view_mode SF
For MF mode, run the following:
python -W ignore eval.py --weight_path ./weights_MF --novel_view_mode MF
For SF mode, run the following:
python -W ignore train.py --novel_view_mode SF
For MF mode, run the following:
python -W ignore train.py --novel_view_mode MF
@inproceedings{tian2025drivingforward,
title={DrivingForward: Feed-forward 3D Gaussian Splatting for Driving Scene Reconstruction from Flexible Surround-view Input},
author={Qijian Tian and Xin Tan and Yuan Xie and Lizhuang Ma},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
year={2025}
}
This implementation is based on the following project:
Dataset-Governance-Policy
PackNet-SfM
gaussian-splatting-jittor
Jittor_Perceptual-Similarity-Metric