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

2017/04/21 by Christopher A. Metzler, Metzler, Christopher A., Alireza Mousavi +3 · 15 citations
Computer Science · Engineering · #Advanced Image Processing Techniques #Blind Source Separation Techniques #FOS: Computer and information sciences #Image and Signal Denoising Methods #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Sparse and Compressive Sensing Techniques

paper · pdf · doi:10.48550/arxiv.1704.06625

openalex publication_date 2017/04/21 · openalex created_date 2019/06/27 · openalex updated_date 2026/07/28

Abstract

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

Cited by

• Score-Based Turbo Message Passing for Plug-and-Play Compressive Imaging
• Time-Embedded Algorithm Unrolling for Computational MRI
• Unsupervised Deep Equilibrium Model Learning for Large-Scale Channel Estimation with Performance Guarantees
• Physics-guided Deep Unfolding Network for Enhanced Kronecker Compressive sensing
• Diffusion Models for Future Networks and Communications: A Comprehensive Survey
• Generative Diffusion Models for Wireless Networks: Fundamental, Architecture, and State-of-the-Art
• Deep Decomposition Learning for Inverse Imaging Problems
• Compressed Sensing via Measurement-Conditional Generative Models
• GAP-net for Snapshot Compressive Imaging
• Deep Learning-based Channel Estimation for Beamspace mmWave Massive MIMO Systems
• Learned Interpretable Residual Extragradient ISTA for Sparse Coding
• SeqCSIST: Sequential Closely-Spaced Infrared Small Target Unmixing
• Feasibility-based Fixed Point Networks
• ISTA-Net++: Flexible Deep Unfolding Network for Compressive Sensing
• Probabilistic Residual Learning for Aleatoric Uncertainty in Image\n Restoration

Related

• Learning to Invert: Signal Recovery via Deep Convolutional Networks
• Compressed Learning: A Deep Neural Network Approach
• Deep Sparse Coding Using Optimized Linear Expansion of Thresholds
• DeepCodec: Adaptive Sensing and Recovery via Deep Convolutional Neural Networks
• Deep Networks for Compressed Image Sensing