base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
datasets:
- earth-insights/EarthReason
library_name: transformers
pipeline_tag: image-segmentation
license: apache-2.0
Bridging Semantics and Geometry: A Decoupled LVLM–SAM Framework for Reasoning Segmentation in Remote Sensing
This repository contains the 7B model of Think2Seg-RS, a decoupled framework for reasoning segmentation in remote sensing (RS) imagery.
The model was introduced in the paper Bridging Semantics and Geometry: A Decoupled LVLM-SAM Framework for Reasoning Segmentation in Remote Sensing.
Overview
Think2Seg-RS decouples high-level semantic reasoning from low-level geometric execution. It trains an LVLM prompter (based on Qwen-2.5-VL) to control a frozen Segment Anything Model (SAM2) via structured geometric prompts. Through a mask-only reinforcement learning objective, the LVLM learns to translate abstract semantic reasoning into spatially grounded actions, achieving state-of-the-art performance on the EarthReason dataset.
Resources
- Paper: arXiv:2512.19302
- Code: GitHub - Think2Seg-RS
- Dataset: EarthReason
Citation
If you find this work helpful for your research, please cite:
@article{think2seg_rs_2025,
title={Bridging Semantics and Geometry: A Decoupled LVLM-SAM Framework for Reasoning Segmentation in Remote Sensing},
author={Anonymous},
journal={arXiv preprint arXiv:2512.19302},
year={2025}
}