Flash-Mono: Feed-Forward Accelerated Gaussian Splatting Monocular SLAM

ICLR 2026 Poster | Project Page | Paper (PDF) | OpenReview

This repository hosts the project page for Flash-Mono.

Authors

Zicheng Zhang¹, Ke Wu¹, Xiangting Meng², Keyu Liu³, Jieru Zhao³, Wenchao Ding¹

¹Fudan University   ²ShanghaiTech University   ³Shanghai Jiao Tong University

Abstract

Monocular 3D Gaussian Splatting SLAM suffers from critical limitations in time efficiency, geometric accuracy, and multi-view consistency. We present Flash-Mono, a system composed of three core modules: a feed-forward prediction frontend, a 2D Gaussian Splatting mapping backend, and an efficient hidden-state-based loop closure module. By directly predicting Gaussian attributes, our method bypasses the burdensome per-frame optimization required in optimization-based GS-SLAM, achieving a 10x speedup while ensuring high-quality rendering.

Citation

@inproceedings{zhang2026flashmono,
  title={Flash-Mono: Feed-Forward Accelerated Gaussian Splatting Monocular {SLAM}},
  author={Zicheng Zhang and Ke Wu and Xiangting Meng and Keyu Liu and Jieru Zhao and Wenchao Ding},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://openreview.net/forum?id=nv3q3crc5D}
}

Acknowledgments

This project page is built using the Academic Project Page Template. Licensed under CC BY-SA 4.0.