SegEarth-R2: Towards Comprehensive Language-guided Segmentation for Remote Sensing Images

• 05/24/2026: The code is released!
• 05/23/2026: LaSeRS dataset is released on HuggingFace

🔧 Usage：

Follow the guidelines below to set up, train and evaluate:

• Preparation ⚙️: Instructions for organizing datasets and pretrained weights for proper model training and inference.
• Installation 💻: Set up the segearthr2 conda environment, install dependencies, and clone the repo.
• Training 🏋️‍♂️: Run scripts/train.sh with DeepSpeed, modifying parameters like data and model paths for training.
• Evaluation 🎯: Run scripts/eval.sh to evaluate the model, updating paths as needed.

⭐️ Citation

If you find this project useful, welcome to cite us.

@inproceedings{xin2026segearth,
  title={Segearth-r2: Towards comprehensive language-guided segmentation for remote sensing images},
  author={Xin, Zepeng and Li, Kaiyu and Chen, Luodi and Li, Wanchen and Yuchen, Xiao and Qiao, Hui and Zhang, Weizhan and Meng, Deyu and Cao, Xiangyong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={13199--13210},
  year={2026}
}