Compressed Sensing with Deep Image Prior and Learned Regularization

Abstract

We propose a novel method for compressed sensing recovery using untrained deep generative models. Our method is based on the recently proposed Deep Image Prior (DIP), wherein the convolutional weights of the network are optimized to match the observed measurements. We show that this approach can be applied to solve any differentiable linear inverse problem, outperforming previous unlearned methods. Unlike various learned approaches based on generative models, our method does not require pre-training over large datasets. We further introduce a novel learned regularization technique, which incorporates prior information on the network weights. This reduces reconstruction error, especially for noisy measurements. Finally, we prove that, using the DIP optimization approach, moderately overparameterized single-layer networks can perfectly fit any signal despite the non-convex nature of the fitting problem. This theoretical result provides justification for early stopping.

References in corpus (22)

• Scikit-learn: Machine Learning in Python
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• WaveNet: A Generative Model for Raw Audio
• Simultaneous analysis of Lasso and Dantzig selector
• Deep Image Prior
• High-dimensional regression with noisy and missing data: Provable guarantees with nonconvexity
• Gradient Descent Provably Optimizes Over-parameterized Neural Networks
• Meta-Learning with Latent Embedding Optimization
• Learned D-AMP: Principled Neural Network based Compressive Image Recovery
• Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks
• Deep Generative Adversarial Networks for Compressed Sensing Automates MRI
• Towards moderate overparameterization: global convergence guarantees for training shallow neural networks
• The lower tail of random quadratic forms, with applications to ordinary least squares and restricted eigenvalue properties
• One Network to Solve Them All --- Solving Linear Inverse Problems using Deep Projection Models
• Overparameterized Nonlinear Learning: Gradient Descent Takes the Shortest Path?
• Modeling Sparse Deviations for Compressed Sensing using Generative Models
• Correction by Projection: Denoising Images with Generative Adversarial Networks
• Recurrent Generative Adversarial Networks for Proximal Learning and Automated Compressive Image Recovery
• Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization
• Solving Linear Inverse Problems Using GAN Priors: An Algorithm with Provable Guarantees
• SUNLayer: Stable denoising with generative networks
• Inference in Deep Networks in High Dimensions

Cited by in corpus (47)

• Deep Image Prior
• Computed Tomography Reconstruction Using Deep Image Prior and Learned Reconstruction Methods
• Learning to Optimize: A Primer and A Benchmark
• 2022 Review of Data-Driven Plasma Science
• Deep Learning Methods for Solving Linear Inverse Problems: Research Directions and Paradigms
• Untrained networks for compressive lensless photography
• Computational Imaging and Artificial Intelligence: The Next Revolution of Mobile Vision
• Algorithmic Guarantees for Inverse Imaging with Untrained Network Priors
• Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators
• Combining Weighted Total Variation and Deep Image Prior for natural and medical image restoration via ADMM
• Measuring Robustness in Deep Learning Based Compressive Sensing
• Generative Models for Low-Rank Video Representation and Reconstruction
• Solving Inverse Problems with Hybrid Deep Image Priors: the challenge of preventing overfitting
• Compressive sensing with un-trained neural networks: Gradient descent finds the smoothest approximation
• Compressible Latent-Space Invertible Networks for Generative Model-Constrained Image Reconstruction
• RDFNet: Regional Dynamic FISTA-Net for Spectral Snapshot Compressive Imaging
• SINR: Deconvolving Circular SAS Images Using Implicit Neural Representations
• Optimal Transport driven CycleGAN for Unsupervised Learning in Inverse Problems
• Learned imaging with constraints and uncertainty quantification
• Understanding Overparameterization in Generative Adversarial Networks
• An untrained deep learning method for reconstructing dynamic magnetic resonance images from accelerated model-based data
• Neural Radiance Flow for 4D View Synthesis and Video Processing
• Regularizing linear inverse problems with convolutional neural networks
• Manifold Modeling in Embedded Space: A Perspective for Interpreting Deep Image Prior
• Deep Learning Techniques for Inverse Problems in Imaging
• A Dataset-free Deep learning Method for Low-Dose CT Image Reconstruction
• Towards Sample-Optimal Compressive Phase Retrieval with Sparse and Generative Priors
• Inference in Multi-Layer Networks with Matrix-Valued Unknowns
• Unsupervised Spatial-spectral Network Learning for Hyperspectral Compressive Snapshot Reconstruction
• Attenuating Random Noise in Seismic Data by a Deep Learning Approach
• CSMCNet: Scalable Video Compressive Sensing Reconstruction with Interpretable Motion Estimation
• Constant-Expansion Suffices for Compressed Sensing with Generative Priors
• Information-Theoretic Lower Bounds for Compressive Sensing with Generative Models
• Solving inverse problems via auto-encoders
• Learning Illumination Patterns for Coded Diffraction Phase Retrieval
• DeepInit Phase Retrieval
• A Projectional Ansatz to Reconstruction
• Deep learning using a biophysical model for Robust and Accelerated Reconstruction (RoAR) of quantitative and artifact-free R2* images
• Provably Convergent Algorithms for Solving Inverse Problems Using Generative Models
• Distributed Coding of Quantized Random Projections
• Robust 1-bit Compressive Sensing with Partial Gaussian Circulant Matrices and Generative Priors
• Variational framework for partially-measured physical system control: examples of vision neuroscience and optical random media
• Beyond Independent Measurements: General Compressed Sensing with GNN Application
• Non-Convex Compressed Sensing with Training Data
• Generative Model Adversarial Training for Deep Compressed Sensing
• Deep Chaos Synchronization
• Compressed Sensing via Measurement-Conditional Generative Models