VARestorer: One-Step VAR Distillation for Real-World Image Super-Resolution

2026-04-23

AI Summary

This work addresses the limitations of existing visual autoregressive models (VARs) in real-world image super-resolution, where causal attention constraints and iterative prediction mechanisms hinder effective use of global context and lead to error accumulation, resulting in blurry outputs. To overcome these issues, the authors propose VARestorer, the first single-step real-image super-resolution framework based on VAR. By leveraging knowledge distillation, a pretrained text-to-image VAR is adapted into an efficient super-resolution model. The approach introduces pyramid image conditioning and cross-scale attention mechanisms, while fine-tuning only 1.2% of the parameters to achieve distribution alignment and multi-scale information fusion. This design substantially mitigates error propagation and accelerates inference by 10× compared to conventional VARs. On DIV2K, VARestorer achieves state-of-the-art performance with 72.32 MUSIQ and 0.7669 CLIPIQA scores.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent advancements in visual autoregressive models (VAR) have demonstrated their effectiveness in image generation, highlighting their potential for real-world image super-resolution (Real-ISR). However, adapting VAR for ISR presents critical challenges. The next-scale prediction mechanism, constrained by causal attention, fails to fully exploit global low-quality (LQ) context, resulting in blurry and inconsistent high-quality (HQ) outputs. Additionally, error accumulation in the iterative prediction severely degrades coherence in ISR task. To address these issues, we propose VARestorer, a simple yet effective distillation framework that transforms a pre-trained text-to-image VAR model into a one-step ISR model. By leveraging distribution matching, our method eliminates the need for iterative refinement, significantly reducing error propagation and inference time. Furthermore, we introduce pyramid image conditioning with cross-scale attention, which enables bidirectional scale-wise interactions and fully utilizes the input image information while adapting to the autoregressive mechanism. This prevents later LQ tokens from being overlooked in the transformer. By fine-tuning only 1.2\% of the model parameters through parameter-efficient adapters, our method maintains the expressive power of the original VAR model while significantly enhancing efficiency. Extensive experiments show that VARestorer achieves state-of-the-art performance with 72.32 MUSIQ and 0.7669 CLIPIQA on DIV2K dataset, while accelerating inference by 10 times compared to conventional VAR inference.

Problem

• visual autoregressive models
• real-world image super-resolution
• causal attention
• error accumulation
• global context

Innovation

• one-step super-resolution
• VAR distillation
• pyramid image conditioning
• cross-scale attention
• parameter-efficient adaptation