InstantIR: Blind Image Restoration with Instant Generative Reference

TL;DR

InstantIR is a diffusion-based blind image restoration method that dynamically uses a reference image and adaptive sampling to handle unknown degradations, achieving state-of-the-art results and enabling creative restoration.

Contribution

The paper introduces InstantIR, a novel diffusion-based approach that adaptively incorporates generative references during inference for improved blind image restoration.

Findings

Achieves state-of-the-art performance on benchmark datasets.

Effectively handles extreme degradations with creative restoration.

Uses variance of references as an indicator for adaptive sampling.

Abstract

Handling test-time unknown degradation is the major challenge in Blind Image Restoration (BIR), necessitating high model generalization. An effective strategy is to incorporate prior knowledge, either from human input or generative model. In this paper, we introduce Instant-reference Image Restoration (InstantIR), a novel diffusion-based BIR method which dynamically adjusts generation condition during inference. We first extract a compact representation of the input via a pre-trained vision encoder. At each generation step, this representation is used to decode current diffusion latent and instantiate it in the generative prior. The degraded image is then encoded with this reference, providing robust generation condition. We observe the variance of generative references fluctuate with degradation intensity, which we further leverage as an indicator for developing a sampling algorithm adaptive to input quality. Extensive experiments demonstrate InstantIR achieves state-of-the-art performance and offering outstanding visual quality. Through modulating generative references with textual description, InstantIR can restore extreme degradation and additionally feature creative restoration.