New model card

README.md

@@ -5,58 +5,111 @@ tags:
   - reflection-removal
   - highlight-removal
   - computer-vision
+  - dinov3
+  - surgical-imaging
 ---
 
-# UnReflectAnything – Pretrained Weights
+# UnReflectAnything: RGB-Only Highlight Removal
 
-Pretrained weights for **UnReflectAnything**: *RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision*.
+[![Project](https://img.shields.io/badge/Project-webpage-ff611b?logo=googlehome&logoColor=ff611b)](https://alberto-rota.github.io/UnReflectAnything/)
+[![PyPI](https://img.shields.io/badge/pip%20install-PyPI-76b1f3?logo=python&logoColor=76b1f3)](https://pypi.org/project/unreflectanything/)
+[![Paper](https://img.shields.io/badge/Paper-arXiv-B31B1B?logo=arxiv&logoColor=B31B1B)](https://arxiv.org/abs/2512.09583)
+[![Weights](https://img.shields.io/badge/Weights-HF%20-FFD21E?logo=huggingface&logoColor=FFD21E)](https://huggingface.co/spaces/AlbeRota/UnReflectAnything)
+[![Demo](https://img.shields.io/badge/Demo-HF%20-FFD21E?logo=huggingface&logoColor=FFD21E)](https://huggingface.co/AlbeRota/UnReflectAnything)
 
-- **Project**: [alberto-rota.github.io/UnReflectAnything](https://alberto-rota.github.io/UnReflectAnything/)
-- **Paper**: [arXiv](https://arxiv.org/abs/2501.12345)
-- **PyPI**: `pip install unreflectanything`
-- **Demo**: [Hugging Face Space](https://huggingface.co/spaces/AlbeRota/UnReflectAnything)
+**UnReflectAnything** is a foundation-model-powered framework that inputs any RGB image and removes specular highlights, returning a clean, diffuse-only output. Unlike traditional methods, it does not require polarization or multi-view data.
 
-## Model description
+---
+
+## Model Capabilities
+* **Domain Versatility**: Works effectively on both natural indoor scenes and specialized surgical/endoscopic domain data.
+* **Synthetic Supervision**: Trained by synthesizing specularities and supervising the reconstruction in the **DINOv3 feature space**.
+* **Foundation Power**: Leverages robust feature representations to generalize highlight removal across diverse lighting conditions.
+
+![Architecture](https://raw.githubusercontent.com/alberto-rota/UnReflectAnything/refs/heads/main/assets/architecture.png)
 
-These weights support the `unreflectanything` package for removing specular highlights and reflections from RGB images without polarization or multi-view input.
+---
+
+## Available Weights
+The following checkpoints are provided for different integration needs:
 
 | File | Description |
-|------|-------------|
-| `fullmodel_896.pt` | Full model checkpoint (recommended for inference) |
-| `decoder_896.pth` | Decoder weights (896px; e.g. for pretraining) |
-| `rgb_decoder.pth` | RGB decoder weights |
-| `rgb_film_decoder.pth` | RGB FiLM decoder weights |
-| `token_inpainter.pth` | Token inpainter module weights |
+| :--- | :--- |
+| **`fullmodel_896.pt`** | Full model checkpoint; recommended for standard inference. |
+| `decoder_896.pth` | Decoder weights optimized for 896px resolution. |
+| `token_inpainter.pth` | Weights for the specialized token inpainter module. |
+| `rgb_decoder.pth` | Standard RGB decoder weights. |
+| `rgb_film_decoder.pth` | RGB FiLM (Feature-wise Linear Modulation) decoder weights. |
+
+---
+
+## Installation & Quick Start
+
+### 1. Install via pip
+```bash
+pip install unreflectanything
+
+```
+
+*Note: Minimum Python 3.11 is required; 3.12 is recommended.*
 
-## How to use
+### 2. Download Weights
 
-1. **Install the package** (with optional weight-download support):
+Use the built-in CLI to fetch the necessary files:
 
-   ```bash
-   pip install unreflectanything[weights]
-   ```
+```bash
+# Alias 'ura' or 'unreflect' also work
+unreflectanything download --weights
 
-2. **Download these weights** into the default cache:
+```
 
-   ```bash
-   unreflectanything download-weights
-   ```
+### 3. Basic Python Usage
 
-   By default the CLI uses the official weights repo. To use this Hugging Face repo instead, set:
+```python
+import unreflectanything
+import torch
 
-   ```bash
-   export UNREFLECTANYTHING_WEIGHTS_REPO="<your-username>/<this-repo-name>"
-   unreflectanything download-weights
-   ```
+# Load the pretrained model (uses cached weights)
+unreflect_model = unreflectanything.model(pretrained=True) 
 
-   Weights are stored under `~/.cache/unreflectanything/weights` (or `$XDG_CACHE_HOME/unreflectanything/weights`). Use `--output-dir` to override.
+# Run inference on a tensor [B, 3, H, W] in range [0, 1]
+images = torch.rand(2, 3, 448, 448).cuda()
+diffuse_output = unreflect_model(images) 
 
-3. **Run inference** with a YAML config that points `weights_path` to the downloaded directory (e.g. to `fullmodel_896.pt` for full-model inference).
+# Simple file-based inference
+unreflectanything.inference("input_with_highlights.png", output="diffuse_result.png")
+
+```
+
+---
+
+## CLI Overview
+
+The package provides a comprehensive command-line interface via `ura`, `unreflect`, or `unreflectanything`.
+
+* **Inference**: `ura inference --input /path/to/images --output /path/to/output`
+* **Evaluation**: `ura evaluate --output /path/to/results --gt /path/to/groundtruth`
+* **Verification**: `ura verify --dataset /path/to/dataset`
+
+---
 
 ## Citation
 
-If you use these weights or the method, please cite the paper and/or link to this repository.
+If you use UnReflectAnything in your research or pipeline, please cite our paper:
+
+```bibtex
+@misc{rota2025unreflectanythingrgbonlyhighlightremoval,
+      title={UnReflectAnything: RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision}, 
+      author={Alberto Rota and Mert Kiray and Mert Asim Karaoglu and Patrick Ruhkamp and Elena De Momi and Nassir Navab and Benjamin Busam},
+      year={2025},
+      eprint={2512.09583},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={[https://arxiv.org/abs/2512.09583](https://arxiv.org/abs/2512.09583)}, 
+}
 
-## License
+```
+
+---
 
-MIT.
+**License**: This project is licensed under the [MIT License](https://mit-license.org/).