---
base_model:
- xinlai/LISA-7B-v1
pipeline_tag: image-text-to-text
---

# ROSALIA-7B-v1

**ROSALIA** is a vision-language model (VLM) designed for precise lesion segmentation in chest X-rays (CXRs). It is a LISA model fine-tuned on the **MIMIC-ILS** dataset, a large-scale instruction-answer dataset for CXR lesion segmentation.

ROSALIA is capable of **Instruction-Guided Lesion Segmentation (ILS)**, a medical-domain adaptation of referring image segmentation (RIS), allowing it to segment diverse lesions and provide textual explanations in response to simple, user-friendly instructions.

This model is the core checkpoint of the paper:  
**[Instruction-Guided Lesion Segmentation for Chest X-rays with Automatically Generated Large-Scale Dataset](https://arxiv.org/abs/2511.15186)**, accepted to **CVPR 2026**.

- **Code:** [https://github.com/checkoneee/ROSALIA](https://github.com/checkoneee/ROSALIA)
- **Dataset:** [https://physionet.org/content/mimic-cxr-ext-ils/1.0.0/](https://physionet.org/content/mimic-cxr-ext-ils/1.0.0/)
- **Paper:** [arXiv:2511.15186](https://arxiv.org/abs/2511.15186)

## 📖 Citation

If you find this model or the related research useful, please cite our work:

```bibtex
@article{choi2025instruction,
  title={Instruction-Guided Lesion Segmentation for Chest X-rays with Automatically Generated Large-Scale Dataset},
  author={Choi, Geon and Yoon, Hangyul and Shin, Hyunju and Park, Hyunki and Seo, Sang Hoon and Yang, Eunho and Choi, Edward},
  journal={arXiv preprint arXiv:2511.15186},
  year={2025}
}
```