Equivariant Sampling for Improving Diffusion Model-based Image Restoration

2025/11/13 by Wu, Chenxu, Kong, Qingpeng, Zhao, Peiang +5
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2511.09965

Abstract

Recent advances in generative models, especially diffusion models, have significantly improved image restoration (IR) performance. However, existing problem-agnostic diffusion model-based image restoration (DMIR) methods face challenges in fully leveraging diffusion priors, resulting in suboptimal performance. In this paper, we address the limitations of current problem-agnostic DMIR methods by analyzing their sampling process and providing effective solutions. We introduce EquS, a DMIR method that imposes equivariant information through dual sampling trajectories. To further boost EquS, we propose the Timestep-Aware Schedule (TAS) and introduce EquS⁺. TAS prioritizes deterministic steps to enhance certainty and sampling efficiency. Extensive experiments on benchmarks demonstrate that our method is compatible with previous problem-agnostic DMIR methods and significantly boosts their performance without increasing computational costs. Our code is available at https://github.com/FouierL/EquS.

Citations

• Diffusion Prior-Based Amortized Variational Inference for Noisy Inverse Problems
• Deep Diffusion Image Prior for Efficient OOD Adaptation in 3D Inverse Problems
• CDFormer:When Degradation Prediction Embraces Diffusion Model for Blind Image Super-Resolution
• Rethinking Diffusion Model for Multi-Contrast MRI Super-Resolution
• Building Bridges across Spatial and Temporal Resolutions: Reference-Based Super-Resolution via Change Priors and Conditional Diffusion Model
• Text-guided Explorable Image Super-resolution
• ID-Blau: Image Deblurring by Implicit Diffusion-based reBLurring AUgmentation
• Diffusion-based Blind Text Image Super-Resolution
• Equivariant plug-and-play image reconstruction
• Deep Equilibrium Diffusion Restoration with Parallel Sampling
• DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior
• Residual Denoising Diffusion Models
• Direct Diffusion Bridge using Data Consistency for Inverse Problems
• Hierarchical Integration Diffusion Model for Realistic Image Deblurring
• Denoising Diffusion Models for Plug-and-Play Image Restoration
• Exploiting Diffusion Prior for Real-World Image Super-Resolution
• Implicit Diffusion Models for Continuous Super-Resolution
• DiffIR: Efficient Diffusion Model for Image Restoration
• Unlimited-Size Diffusion Restoration
• I²SB: Image-to-Image Schrödinger Bridge
• DifFace: Blind Face Restoration with Diffused Error Contraction
• CiaoSR: Continuous Implicit Attention-in-Attention Network for Arbitrary-Scale Image Super-Resolution
• Multiscale Structure Guided Diffusion for Image Deblurring
• Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model
• GAN Prior based Null-Space Learning for Consistent Super-Resolution
• Solving 3D Inverse Problems using Pre-trained 2D Diffusion Models
• Diffusion Posterior Sampling for General Noisy Inverse Problems
• Lossy Image Compression with Conditional Diffusion Models
• Imaging with Equivariant Deep Learning
• Reference-based Image Super-Resolution with Deformable Attention Transformer
• Towards Interpretable Video Super-Resolution via Alternating Optimization
• Elucidating the Design Space of Diffusion-Based Generative Models
• Simple Baselines for Image Restoration
• MAT: Mask-Aware Transformer for Large Hole Image Inpainting
• VRT: A Video Restoration Transformer
• Denoising Diffusion Restoration Models
• RePaint: Inpainting using Denoising Diffusion Probabilistic Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• Deblurring via Stochastic Refinement
• Robust Equivariant Imaging: a fully unsupervised framework for learning to image from noisy and partial measurements
• Restormer: Efficient Transformer for High-Resolution Image Restoration
• Palette: Image-to-Image Diffusion Models
• ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
• SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations
• Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data
• Noise2Score: Tweedie's Approach to Self-Supervised Image Denoising\n without Clean Images
• Video Super-Resolution Transformer
• SNIPS: Solving Noisy Inverse Problems Stochastically
• Diffusion Models Beat GANs on Image Synthesis
• SRDiff: Single Image Super-Resolution with Diffusion Probabilistic Models
• Image Super-Resolution via Iterative Refinement
• Aggregated Contextual Transformations for High-Resolution Image Inpainting
• Equivariant Imaging: Learning Beyond the Range Space
• Designing a Practical Degradation Model for Deep Blind Image\n Super-Resolution
• Score-Based Generative Modeling through Stochastic Differential Equations
• Denoising Diffusion Implicit Models
• Plug-and-Play Image Restoration with Deep Denoiser Prior
• Improving Generative Adversarial Networks with Local Coordinate Coding
• Denoising Diffusion Probabilistic Models
• Exploiting Deep Generative Prior for Versatile Image Restoration and Manipulation
• PULSE: Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models
• DuDoRNet: Learning a Dual-Domain Recurrent Network for Fast MRI Reconstruction with Deep T1 Prior
• Noisier2Noise: Learning to Denoise from Unpaired Noisy Data
• Image Inpainting with Learnable Bidirectional Attention Maps
• Generative Modeling by Estimating Gradients of the Data Distribution
• DuDoNet: Dual Domain Network for CT Metal Artifact Reduction
• Noise2Self: Blind Denoising by Self-Supervision
• Noise2Void - Learning Denoising from Single Noisy Images
• ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks
• Deep null space learning for inverse problems: convergence analysis and rates
• Adversarial Learning with Local Coordinate Coding
• Free-Form Image Inpainting with Gated Convolution
• Noise2Noise: Learning Image Restoration without Clean Data
• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
• The Perception-Distortion Tradeoff
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Improved Training of Wasserstein GANs
• Accurate Image Super-Resolution Using Very Deep Convolutional Networks
• Compression Artifacts Reduction by a Deep Convolutional Network
• Deep Unsupervised Learning using Nonequilibrium Thermodynamics
• Image Super-Resolution Using Deep Convolutional Networks
• Image Super-Resolution Using Deep Convolutional Networks
• Tweedie’s Formula and Selection Bias
• A Connection Between Score Matching and Denoising Autoencoders
• Bayesian-Based Iterative Method of Image Restoration*

Related

• Deep Equilibrium Diffusion Restoration with Parallel Sampling
• DGSolver: Diffusion Generalist Solver with Universal Posterior Sampling for Image Restoration
• Decoupled Data Consistency with Diffusion Purification for Image Restoration
• EquiReg: Equivariance Regularized Diffusion for Inverse Problems
• Efficient Diffusion Model for Image Restoration by Residual Shifting