Add comprehensive model card for UNICE

README.md

@@ -0,0 +1,69 @@
+---
+license: mit
+pipeline_tag: image-to-image
+library_name: diffusers
+---
+
+# UNICE: Training A Universal Image Contrast Enhancer
+
+This repository contains the **UNICE** model, a Universal Image Contrast Enhancer, as presented in the paper [UNICE: Training A Universal Image Contrast Enhancer](https://huggingface.co/papers/2507.17157).
+
+<p align="center">
+    <a href="https://colab.research.google.com/drive/1EjIAThdFhyE_51ujdAUK0_4NRlBcKIdf?usp=sharing" target="_blank">
+    <img src="https://img.shields.io/badge/Colab%20Demo-F9AB00?style=flat&logo=googlecolab&logoColor=white">
+    </a>
+    &nbsp;&nbsp;&nbsp;
+    <a href="https://huggingface.co/datasets/lahaina/UNICE" target="_blank">
+    <img src="https://img.shields.io/badge/Hugging%20Face-EA6B66?style=flat&logo=huggingface&logoColor=FFD21E">
+    </a>
+    &nbsp;&nbsp;&nbsp;
+    <a href="https://arxiv.org/abs/2507.17157" target="_blank">
+    <img src="https://img.shields.io/badge/arXiv-2507.17157-b31b1b?style=flat&logo=arXiv&logoColor=white">
+    </a>
+    &nbsp;&nbsp;&nbsp;
+    <a href="https://github.com/RuodaiCui/UNICE" target="_blank">
+    <img src="https://img.shields.io/badge/GitHub-Code-blue.svg?logo=github&">
+    </a>
+</p>
+
+## Overview
+Existing image contrast enhancement methods are typically designed for specific tasks. UNICE introduces a novel approach to image contrast enhancement by training a universal and generalized model capable of handling various tasks such as under-/over-exposure correction and low-light enhancement. It operates by first generating a multi-exposure sequence (MES) from a single sRGB image, and then fusing the generated MES into an enhanced image. This method is free of costly human labeling and demonstrates superior generalization performance over existing methods, even outperforming manually created ground-truths in multiple no-reference image quality metrics.
+
+<img src="https://huggingface.co/datasets/lahaina/UNICE/resolve/main/img/GT_unice_cmp.jpg" alt="Comparison with GT" width="600">
+
+## Usage
+For detailed installation instructions, training procedures, and more examples, please refer to the [official GitHub repository](https://github.com/RuodaiCui/UNICE).
+
+Pre-trained weights for this model are available at [Hugging Face](https://huggingface.co/lahaina/unice/tree/main/checkpoints).
+
+To test the model with different exposure values, use the following script:
+
+```bash
+#!/bin/bash
+
+# Define the exposure value
+exposure=0.5
+output_dir="output/$exposure"
+
+CUDA_VISIBLE_DEVICES=5 ../miniconda3/envs/img2img-turbo/bin/python src/inference.py \
+--model_path "checkpoints/exposure.pkl" \
+--input_dir /local/mnt/workspace/ruodcui/code/adaptive_3dlut/data/BAID512/input/ \
+--output_dir $output_dir \
+--prompt "exposure control" \
+--exposure $exposure
+```
+
+## Citation
+If you find our work helpful or inspiring, please consider citing our paper:
+
+```bibtex
+@misc{ruodai2025UNICE,
+      title={UNICE: Training A Universal Image Contrast Enhancer},
+      author={Ruodai Cui and Lei Zhang},
+      year={2025},
+      eprint={2507.17157},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.17157},
+}
+```