A Draft–Verifier speculative sampling framework for accelerating diffusion-based super-resolution

Yipeng Ju

doi:10.1016/j.jvcir.2026.104998

A Draft–Verifier speculative sampling framework for accelerating diffusion-based super-resolution

2026-10-11

0

PRE

AI

DOI：10.1016/j.jvcir.2026.104998

Original

Save

Abstract

En 中文

• We adapt speculative sampling to continuous-latent diffusion-based super-resolution using a Draft–Verifier framework. • A lightweight draft model proposes candidate denoising updates, while an intermediate-representation verifier controls acceptance and fallback. • Experiments under DDIM 20-step sampling demonstrate measurable acceleration and reveal practical quality-efficiency trade-offs.

Keywords:

Diffusion super-resolution

Speculative sampling

Draft–verifier framework

Efficient inference

Image restoration

A Draft–Verifier speculative sampling framework for accelerating diffusion-based super-resolution

Abstract

Journal

Organization

Cited Papers