Add model card and metadata

README.md

@@ -1,3 +1,38 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+pipeline_tag: image-to-image
+---
+
+# UAE: Unified Autoencoding
+
+This repository contains the weights for **Unified Autoencoding (UAE)**, introduced in the paper [The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding](https://huggingface.co/papers/2512.19693).
+
+UAE is a model that harmonizes semantic structure and pixel-level details through an innovative frequency-band modulator. By leveraging the "Prism Hypothesis," the model unifies semantic abstraction (low-frequency) and pixel-level fidelity (high-frequency) into a single latent space, achieving state-of-the-art performance on ImageNet and MS-COCO benchmarks.
+
+## Resources
+
+- **Paper:** [The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding](https://huggingface.co/papers/2512.19693)
+- **GitHub Repository:** [https://github.com/WeichenFan/UAE](https://github.com/WeichenFan/UAE)
+
+## Evaluation Results
+
+As reported in the official repository, the model achieves the following performance with a frequency ratio of 1.0:
+
+| Dataset | PSNR | SSIM | rFID |
+|---------|------|------|------|
+| ImageNet | 29.588 dB | 0.8789 | 0.193 |
+| MS-COCO | 29.484 dB | 0.8846 | 0.157 |
+
+## Citation
+
+```bibtex
+@misc{fan2025uae,
+      title={The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding}, 
+      author={Weichen Fan and Haiwen Diao and Quan Wang and Dahua Lin and Ziwei Liu},
+      year={2025},
+      eprint={2512.19693},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2512.19693}, 
+}
+```