Projecting VAE latents to a fixed spherical radius and replacing linear interpolation with spherical linear interpolation improves class-conditional ImageNet-256 FID while leaving the diffusion architecture unchanged.
Riemannian diffusion models.Advances in Neural Information Processing Systems, 35:2750– 2761, 2022
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Aligning Latent Geometry for Spherical Flow Matching in Image Generation
Projecting VAE latents to a fixed spherical radius and replacing linear interpolation with spherical linear interpolation improves class-conditional ImageNet-256 FID while leaving the diffusion architecture unchanged.