Learned D-AMP: Principled Neural Network based Compressive Image Recovery

Abstract

Compressive image recovery is a challenging problem that requires fast and accurate algorithms. Recently, neural networks have been applied to this problem with promising results. By exploiting massively parallel GPU processing architectures and oodles of training data, they can run orders of magnitude faster than existing techniques. However, these methods are largely unprincipled black boxes that are difficult to train and often-times specific to a single measurement matrix. It was recently demonstrated that iterative sparse-signal-recovery algorithms can be "unrolled" to form interpretable deep networks. Taking inspiration from this work, we develop a novel neural network architecture that mimics the behavior of the denoising-based approximate message passing (D-AMP) algorithm. We call this new network Learned D-AMP (LDAMP). The LDAMP network is easy to train, can be applied to a variety of different measurement matrices, and comes with a state-evolution heuristic that accurately predicts its performance. Most importantly, it outperforms the state-of-the-art BM3D-AMP and NLR-CS algorithms in terms of both accuracy and run time. At high resolutions, and when used with sensing matrices that have fast implementations, LDAMP runs over $50\times$ faster than BM3D-AMP and hundreds of times faster than NLR-CS.

Cited by in corpus (44)

• Compressed Sensing with Deep Image Prior and Learned Regularization
• Multi-Channel Deep Networks for Block-Based Image Compressive Sensing
• Learning to Optimize: A Primer and A Benchmark
• Deep Learning Methods for Solving Linear Inverse Problems: Research Directions and Paradigms
• GAP-net for Snapshot Compressive Imaging
• Small Sample Learning in Big Data Era
• Unsupervised Learning with Stein's Unbiased Risk Estimator
• Compressed Domain Image Classification Using a Dynamic-Rate Neural Network
• Deep Unfolding with Normalizing Flow Priors for Inverse Problems
• Multi-Layer Bilinear Generalized Approximate Message Passing
• Learning Space Partitions for Nearest Neighbor Search
• Adaptive Block Compressive Sensing: towards a real-time and low-complexity implementation
• Deep Learning-based Channel Estimation for Beamspace mmWave Massive MIMO Systems
• Neumann Networks for Inverse Problems in Imaging
• Learned Robust PCA: A Scalable Deep Unfolding Approach for High-Dimensional Outlier Detection
• Deep Learning Techniques for Inverse Problems in Imaging
• Theoretical Interpretation of Learned Step Size in Deep-Unfolded Gradient Descent
• FlowStep3D: Model Unrolling for Self-Supervised Scene Flow Estimation
• ISTA-Net++: Flexible Deep Unfolding Network for Compressive Sensing
• Deep Learning in Physical Layer Communications
• Probabilistic Residual Learning for Aleatoric Uncertainty in Image Restoration
• PgNN: Physics-guided Neural Network for Fourier Ptychographic Microscopy
• Degrees of Freedom Analysis of Unrolled Neural Networks
• TFPnP: Tuning-free Plug-and-Play Proximal Algorithm with Applications to Inverse Imaging Problems
• Safeguarded Learned Convex Optimization
• Robust Equivariant Imaging: a fully unsupervised framework for learning to image from noisy and partial measurements
• Deep Learning Techniques for Compressive Sensing-Based Reconstruction and Inference -- A Ubiquitous Systems Perspective
• Feasibility-based Fixed Point Networks
• A Flexible Convolutional Solver with Application to Photorealistic Style Transfer
• Phase Retrieval using Expectation Consistent Signal Recovery Algorithm based on Hypernetwork
• Scalable Deep Compressive Sensing
• Deep Decomposition Learning for Inverse Imaging Problems
• Learned Interpretable Residual Extragradient ISTA for Sparse Coding
• Multi-Scale Deep Compressive Imaging
• Deep-Learned Approximate Message Passing for Asynchronous Massive Connectivity
• Bayesian Convolutional Neural Networks for Compressed Sensing Restoration
• Estimation for High-Dimensional Multi-Layer Generalized Linear Model -- Part I: The Exact MMSE Estimator
• OpenICS: Open Image Compressive Sensing Toolbox and Benchmark
• Compressed Sensing via Measurement-Conditional Generative Models
• AMPA-Net: Optimization-Inspired Attention Neural Network for Deep Compressed Sensing
• Distributed Video Adaptive Block Compressive Sensing
• Tree Search Network for Sparse Regression
• Across-domains transferability of Deep-RED in de-noising and compressive sensing recovery of seismic data
• SUREMap: Predicting Uncertainty in CNN-based Image Reconstruction Using Stein's Unbiased Risk Estimate