JTRNEO/RS3DAda

SynRS3D: A Synthetic Dataset for Global 3D Semantic Understanding from Monocular Remote Sensing Imagery

Authors:
Jian Song^1,2, Hongruixuan Chen¹, Weihao Xuan^1,2, Junshi Xia², Naoto Yokoya^1,2

¹ The University of Tokyo
² RIKEN AIP

Conference: Neural Information Processing Systems (Spotlight), 2024

For more details, please refer to our paper and visit our GitHub repository.

---

Overview

TL;DR:
We are excited to release two high-performing models for height estimation and land cover mapping. These models were trained on the SynRS3D dataset using our novel domain adaptation method, RS3DAda.

• Encoder: Vision Transformer (ViT-L), pretrained with DINOv2
• Decoder: DPT, trained from scratch

These models excel in tasks involving large-scale global 3D semantic understanding from high-resolution remote sensing imagery. Feel free to integrate them into your projects for enhanced performance in related applications.

---

How to Cite

If you find the RS3DAda model useful in your research, please consider citing:

@article{song2024synrs3d,
title={SynRS3D: A Synthetic Dataset for Global 3D Semantic Understanding from Monocular Remote Sensing Imagery},
author={Song, Jian and Chen, Hongruixuan and Xuan, Weihao and Xia, Junshi and Yokoya, Naoto},
journal={arXiv preprint arXiv:2406.18151},
year={2024}
}

---

Contact

For any questions or feedback, please reach out via email at song@ms.k.u-tokyo.ac.jp.

We hope you enjoy using the pretrained RS3DAda models!