Year: 2025
Type: preprint
Abstract: Visual Autoregressive Models (VAR) offer efficient and high-quality image generation but suffer from computational redundancy due to repeated Transformer calls at increasing resolutions. We introduce a dynamic Mixture-of-Experts router integrated into VAR. The new architecture allows to trade compute for quality through scale-aware thresholding. This thresholding strategy balances expert selection based on token complexity and resolution, without requiring additional training. As a result, we achieve 20% fewer FLOPs, 11% faster inference and match the image quality achieved by the dense baseline.
Source: arXiv (Cornell University)
Authors: Vincenti, Jort, Jazbec, Metod, Xia, Guoxuan
Institutions: University of Amsterdam, Imperial College London
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4417114134
Cites:
Cited by:
Related to: