README.md

---
license: cc-by-4.0
tags:
- multimodal-retrieval
- text-video-retrieval
- audio-visual
- hypergraph
---

# HyVol: Hypergraph-Regularized Gramian Volumes for Multimodal Retrieval

Pretrained weights for the WACV 2027 paper
**Hypergraph-Regularized Gramian Volumes for Multimodal Retrieval**
by Anindya Nag, Ambuj Mehrish and Sebastiano Vascon.

Paper: [arXiv:2609.15320](https://arxiv.org/abs/2609.15320)

HyVol is a training-only module for volume-based multimodal retrieval. A hypergraph
adds semantic relations between training samples before the volume loss is computed.
The hypergraph is removed after training, so the backbone, scoring function and
retrieval cost are unchanged.

## Checkpoints

| File | Backbone |
|---|---|
| `GRAM_HYVOL/HyVol_GRAM.pt` | GRAM + HyVol |
| `HyperGRAM_HYVOL/HyVol_HyperGRAM.pt` | Hyperbolic GRAM + HyVol |

Both models are trained on a 150K-clip subset of VAST-27M and evaluated zero-shot.
Load each checkpoint with the codebase of its backbone (GRAM or Hyperbolic GRAM).

## Citation

```bibtex
@inproceedings{nag2027hyvol,
  title     = {Hypergraph-Regularized Gramian Volumes for Multimodal Retrieval},
  author    = {Nag, Anindya and Mehrish, Ambuj and Vascon, Sebastiano},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year      = {2027}
}
```