Summary
This arXiv paper (2605.15193) proposes aligning latent geometry for flow matching in image generation. The authors observe that both Gaussian noise and VAE latents concentrate in thin spherical shells, yet standard latent flow matching transports samples along Euclidean chords that leave these shells. Component-swap probes show that decoded perceptual and semantic content is carried predominantly by the angular (directional) component of each latent token, with radius contributing far less. Building on this, the method projects data latents onto a fixed token radius, uses the radial projection of Gaussian noise as a spherical prior, finetunes the decoder with the encoder frozen, and replaces linear interpolation with spherical linear interpolation, yielding geodesic paths that stay on the sphere. The work, by Meral, Oktay, Yesiltepe, Akan, and Yanardag, was posted in May 2026 and falls under computer vision.
Paper Overview
Field: Computer Vision (CV)
Authors: Tuna Han Salih Meral, Kaan Oktay, Hidir Yesiltepe, Adil Kaan Akan, Pinar Yanardag
Posted: 2026-05-14
arXiv: 2605.15193
Abstract (translated)
Latent flow matching for image generation usually transports Gaussian noise to variational autoencoder latents along linear paths. Both endpoints, however, concentrate in thin spherical shells, and a Euclidean chord leaves those shells even when preprocessing aligns their radii. By decomposing each latent token into radial and angular components, we show through component-swap probes that decoded perceptual and semantic content is carried predominantly by direction, with radius contributing much less. We therefore project data latents onto a fixed token radius, use the radial projection of Gaussian noise as the spherical prior, finetune the decoder with the encoder frozen, and replace linear interpolation with spherical linear interpolation. The resulting geodesic paths stay on the sphere.
*(Abstract truncated in source; see the arXiv page for the full text.)*
---
*Auto-collected on 2026-05-15*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177620060