Repository files navigation

Joint effort by Shanghai Innovation Institute (SII) and OpenDriveLab at The University of Hong Kong.

• MTGS leveragesmulti-traversal data for scene reconstruction with better geometry.
• We conduct a robust pipeline to calibrate and reconstruct thenuPlan dataset with multi-traversal data, which is widely used in the autonomous driving community. See downstream applications inNAVSIM v2.
• We integrate aweb viewer from nerfstudio to visualize the reconstructed scene and switch nodes between different traversals.
• Getting started with our codebase now! 🚀

All the videos below are reconstructed and rendered with our method, MTGS, fromroad_block-331220_4690660_331190_4690710.

Rendered results on training traversals 1, 2, and 3, from top to bottom.

Novel-view results on the testing traversal.

• [2025/05/29] We release the checkpoints.Check it out!
• [2025/05/27] Official code release.
• [2025/05/14] Video demo release.
• [2025/03/16] We released ourpaper on arXiv.

• Official code release.
• Release the checkpoints.
• Demo page.

• 📦Installation
• 📊Prepare Data
• 🚀Running

If any parts of our paper and code help your research, please consider citing us and giving a star to our repository.

@article{li2025mtgs,title={MTGS: Multi-Traversal Gaussian Splatting},author={Li, Tianyu and Qiu, Yihang and Wu, Zhenhua and Lindstr{\"o}m, Carl and Su, Peng and Nie{\ss}ner, Matthias and Li, Hongyang},journal={arXiv preprint arXiv:2503.12552},year={2025}}

All content in this repository is under theApache-2.0 license.The released data is based onnuPlan and are under theCC-BY-NC-SA 4.0 license.

We acknowledge all the open-source contributors for the following projects to make this work possible:

• nerfstudio
• gsplat
• drivestudio
• kiss-icp
• UniDepth