| --- |
| library_name: diffusers |
| pipeline_tag: text-to-image |
| --- |
| |
| # Sana 0.6B (CheXGenBench) |
|
|
| This repository contains the Sana 0.6B model checkpoint, which was identified as a top-performing architecture for synthetic chest radiograph generation in the paper [CheXGenBench: A Unified Benchmark For Fidelity, Privacy and Utility of Synthetic Chest Radiographs](https://huggingface.co/papers/2505.10496). |
|
|
| CheXGenBench is a rigorous and multifaceted evaluation framework that assesses synthetic chest radiograph generation across fidelity, privacy risks, and clinical utility. Sana 0.6B achieved state-of-the-art results on this benchmark and was used to generate the **SynthCheX-75K** dataset. |
|
|
| - **Paper:** [CheXGenBench: A Unified Benchmark For Fidelity, Privacy and Utility of Synthetic Chest Radiographs](https://huggingface.co/papers/2505.10496) |
| - **Project Page:** [https://raman1121.github.io/CheXGenBench/](https://raman1121.github.io/CheXGenBench/) |
| - **Repository:** [https://github.com/Raman1121/CheXGenBench](https://github.com/Raman1121/CheXGenBench) |
|
|
| ## Model Description |
| Sana 0.6B is a text-to-image generative model capable of producing high-fidelity medical imagery. In the context of the CheXGenBench benchmark, it demonstrated superior performance in capturing clinical details while maintaining a balance between generation quality and privacy. |
|
|
| For detailed instructions on environment setup, generating synthetic data, and running evaluation metrics (FID, privacy, and clinical utility), please refer to the official [GitHub repository](https://github.com/Raman1121/CheXGenBench). |
|
|
| ## Citation |
| If you find this model or the benchmark useful in your research, please cite: |
|
|
| ```bibtex |
| @article{dutt2025chexgenbench, |
| title={CheXGenBench: A Unified Benchmark For Fidelity, Privacy and Utility of Synthetic Chest Radiographs}, |
| author={Dutt, Raman and Sanchez, Pedro and Yao, Yongchen and McDonagh, Steven and Tsaftaris, Sotirios A and Hospedales, Timothy}, |
| journal={arXiv preprint arXiv:2505.10496}, |
| year={2025} |
| } |
| ``` |
