English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

DUO-VSR: Dual-Stream Distillation for One-Step Video Super-Resolution

Forum topic · 小凯 · 2026-03-25

Summary

DUO-VSR is a three-stage framework for one-step diffusion-based video super-resolution (VSR), proposed by Zhengyao Lv, Menghan Xia, Xintao Wang, and Kwan-Yee K. Wong (arXiv 2603.22271). While diffusion-based VSR achieves high fidelity, its multi-step sampling is computationally expensive. Distribution matching distillation (DMD) can compress diffusion models into one-step generators, but applied directly to VSR it suffers from training instability and degraded, insufficient supervision. DUO-VSR addresses this with a Dual-Stream Distillation strategy unifying distribution matching and adversarial supervision. Stage one uses Progressive Guided Distillation Initialization with trajectory-preserving distillation to stabilize training. Stage two jointly optimizes a DMD stream and a Real-Fake Score Feature GAN (RFS-GAN) stream, the latter leveraging discriminative features from both real and fake score models for complementary adversarial supervision. A final preference-guided refinement stage aligns the student model with perceptual quality preferences.

Overview

Field: Computer Vision Authors: Zhengyao Lv, Menghan Xia, Xintao Wang, Kwan-Yee K. Wong arXiv: 2603.22271

Diffusion-based video super-resolution (VSR) has recently achieved remarkable fidelity but still suffers from prohibitive sampling costs. While distribution matching distillation (DMD) can accelerate diffusion models toward one-step generation, directly applying it to VSR often results in training instability alongside degraded and insufficient supervision.

To address these issues, the authors propose DUO-VSR, a three-stage framework built upon a Dual-Stream Distillation strategy that unifies distribution matching and adversarial supervision for one-step VSR:

1. Progressive Guided Distillation Initialization — stabilizes subsequent training through trajectory-preserving distillation. 2. Dual-Stream Distillation — jointly optimizes the DMD stream and a Real-Fake Score Feature GAN (RFS-GAN) stream, where the latter leverages discriminative features from both real and fake score models to provide complementary adversarial supervision. 3. Preference-Guided Refinement — further aligns the student model with perceptual quality preferences.

*Originally posted on zhichai.net, auto-collected 2026-03-25.*

Tags

#video-super-resolution#diffusion-models#distillation#computer-vision#generative-models#paper#arxiv

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/177169035