Add model card, pipeline tag and library metadata

README.md

@@ -1,3 +1,41 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+pipeline_tag: image-segmentation
+library_name: transformers
+---
+
+# UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes
+
+UniGeoSeg is a unified framework for open-world segmentation in geospatial scenes. It is designed to handle various instruction-driven segmentation tasks in remote sensing, including referring, interactive, and reasoning-based segmentation.
+
+- **Paper:** [UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes](https://huggingface.co/papers/2511.23332)
+- **Repository:** [GitHub - MiliLab/UniGeoSeg](https://github.com/MiliLab/UniGeoSeg)
+- **Benchmark:** [GeoSeg-Bench](https://huggingface.co/datasets/nishuo1999/GeoSeg-Bench)
+
+## Introduction
+UniGeoSeg addresses the challenges of fragmented task formulations and limited instruction data in remote sensing by leveraging a progressive training strategy and a unified architecture. It is trained on the **GeoSeg-1M** dataset, which contains 1.1 million image-mask-instruction triplets, providing strong zero-shot generalization capabilities across complex geospatial scenes.
+
+The model incorporates task-aware text enhancement and latent knowledge memory to facilitate multi-task learning, achieving state-of-the-art performance across diverse geospatial benchmarks.
+
+## Usage
+
+For inference and evaluation, please refer to the scripts provided in the official repository. You can run the provided inference script using:
+
+```bash
+python scripts/eval.sh
+```
+
+## Citation
+If you find UniGeoSeg helpful, please cite the following work:
+
+```bibtex
+@misc{ni2025unigeosegunifiedopenworldsegmentation,
+      title={UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes}, 
+      author={Shuo Ni and Di Wang and He Chen and Haonan Guo and Ning Zhang and Jing Zhang},
+      year={2025},
+      eprint={2511.23332},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2511.23332}, 
+}
+```