Although the prior is trained only on human shapes, our method generalizes to unseen categories, enabling robust shape matching across domains! (5/n)

This learned spectral diffusion prior is then distilled and integrated into deep functional map frameworks. (4/n)

We train a diffusion model directly on functional maps to learn their underlying structures as a category-agnostic spectral prior. (3/n)

Functional maps provide a spectral representation of pointwise correspondences between shapes.
Interestingly, they exhibit similar diagonal structures across humans, animals, and other categories. (2/n)

📢 Excited to share our latest #ICCV2025 work DiffuMatch: learning spectral diffusion priors for robust non-rigid shape matching! (1/n)

Je l'insulte régulièrement, il me répond de manière passive agressive et ça m'énerve encore plus