---
task_categories:
- image-to-text
tags:
- medical
- radiology
- chest-xray
---

# REVA-PO Datasets

This repository contains the processed annotations and dataset structures used in the paper **REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation** (ECCV 2026).

[Paper (arXiv)](https://arxiv.org/abs/2607.10147) | [Code (GitHub)](https://github.com/LiGuo12/REVA_PO/)

## Dataset Structure

### 1. IU-Xray
Download the [IU-Xray](https://huggingface.co/datasets/liguo12/REVA_PO_Datasets/tree/main/iuxray_dataset) files and unzip `images.zip`. 

After unzipping, the directory structure should look like this:
```text
iuxray_dataset/
├── images/
└── annotation_with_categories.json
```

### 2. MIMIC-CXR
Download the chest X-ray images of the MIMIC-CXR dataset from the [official PhysioNet website](https://physionet.org/content/mimic-cxr-jpg/2.0.0/), and download the [annotations](https://huggingface.co/datasets/liguo12/REVA_PO_Datasets/tree/main/mimic_dataset) from this repository.

Place the annotations and images (`files` directory from PhysioNet) into the same folder:
```text
mimic_dataset/
├── files/
├── mimic_with_categories.json
└── mimic_with_categories_sampled_10k.json
```

## Citation

If you find this work or the datasets helpful, please cite the paper:

```bibtex
@misc{guo2026revapo,
      title={REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation}, 
      author={Li Guo and Anas M. Tahir and Z. Jane Wang},
      year={2026},
      eprint={2607.10147},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2607.10147}, 
}
```