# HybridMQA Checkpoint

This repository hosts the model weights for **HybridMQA**, a full-reference quality assessment method for 3D colored meshes. HybridMQA leverages both geometry and texture representations and models their interactions to predict perceptual quality scores.

📄 Paper: [arXiv:2412.01986](https://arxiv.org/abs/2412.01986)  
🌐 Project Page: [https://arshafiee.github.io/hybridmqa/](https://arshafiee.github.io/hybridmqa/)  
💻 Codebase: [GitHub Repository](https://github.com/arshafiee/hybridmqa)

## Checkpoints

This repository includes three trained HybridMQA checkpoints:

| Filename                      | Dataset  | kfold_seed       | shuffle_seed     | Notes                                   |
|-------------------------------|----------|------------------|------------------|-----------------------------------------|
| `ckpt_TMQA_adapted.pth`       | TMQA     | 27               | 0                | Trained on 80% with the given seeds     |
| `ckpt_TSMD_adapted.pth`       | TSMD     | 7                | 2                | Trained on 80% with the given seeds     |
| `ckpt_YN2023_adapted.pth`     | YN2023   | 7                | 1                | Trained on 80% with the given seeds     |

## Usage

Download the checkpoint using:

```python
from huggingface_hub import hf_hub_download

ckpt_path = hf_hub_download(
    repo_id="arshafiee/hybridmqa-checkpoint",
    filename="<checkpoint_name>.pth"
)
```


## Citation

If you use this work in your research, please cite:

```bibtex
@InProceedings{sarvestani2025hybridmqa,
    author    = {Sarvestani, Armin Shafiee and Tang, Sheyang and Wang, Zhou},
    title     = {HybridMQA: Exploring Geometry-Texture Interactions for Colored Mesh Quality Assessment},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    year      = {2025},
    pages     = {21414-21424}
}
```

## License

This model is released under the MIT License.