---
title: UNCHA Hyperbolic VLM
emoji: 🔮
colorFrom: pink
colorTo: gray
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
short_description: Hyperbolic vision-language zero-shot classification
python_version: "3.12"
startup_duration_timeout: 30m
---

# UNCHA: Uncertainty-guided Compositional Hyperbolic Alignment

This Space demonstrates **zero-shot image classification** using the UNCHA model,
a hyperbolic vision-language model that embeds images and text in Lorentz (hyperbolic)
space to capture hierarchical part-whole relationships.

## How it works

1. Upload an image (or use an example)
2. Provide a comma-separated list of candidate labels
3. The model encodes the image and each label's text prompts into hyperbolic space
4. Classification scores are computed via the Lorentzian inner product

## References

- [Paper (arXiv)](https://arxiv.org/abs/2603.22042)
- [GitHub](https://github.com/jeeit17/UNCHA)
- [Model weights](https://huggingface.co/hayeonkim/uncha)