Purrception: Variational Flow Matching for Vector-Quantized Image Generation

MCML Authors

Vincent Tao Hu

Dr.

→ Group Björn Ommer
 Computer Vision & Learning

Björn Ommer

Prof. Dr.

Principal Investigator

Computer Vision & Learning

Abstract

We introduce Purrception, a variational flow matching approach for vector-quantized image generation that provides explicit categorical supervision while maintaining continuous transport dynamics. Our method adapts Variational Flow Matching to vector-quantized latents by learning categorical posteriors over codebook indices while computing velocity fields in the continuous embedding space. This combines the geometric awareness of continuous methods with the discrete supervision of categorical approaches, enabling uncertainty quantification over plausible codes and temperature-controlled generation. We evaluate Purrception on ImageNet-1k 256x256 generation. Training converges faster than both continuous flow matching and discrete flow matching baselines while achieving competitive FID scores with state-of-the-art models. This demonstrates that Variational Flow Matching can effectively bridge continuous transport and discrete supervision for improved training efficiency in image generation.

misc

---

Preprint

Oct. 2025

---

Authors

R.-A. Matişan • V. T. Hu • G. Bartosh • B. Ommer • C. G. M. Snoek • M. Welling • J.-W. van de Meent • M. M. Derakhshani • F. Eijkelboom

---

Links

---

Research Area

 B1 | Computer Vision