TFPnP: Tuning-free Plug-and-Play Proximal Algorithm with Applications to Inverse Imaging Problems

Abstract

Plug-and-Play (PnP) is a non-convex optimization framework that combines proximal algorithms, for example, the alternating direction method of multipliers (ADMM), with advanced denoising priors. Over the past few years, great empirical success has been obtained by PnP algorithms, especially for the ones that integrate deep learning-based denoisers. However, a key challenge of PnP approaches is the need for manual parameter tweaking as it is essential to obtain high-quality results across the high discrepancy in imaging conditions and varying scene content. In this work, we present a class of tuning-free PnP proximal algorithms that can determine parameters such as denoising strength, termination time, and other optimization-specific parameters automatically. A core part of our approach is a policy network for automated parameter search which can be effectively learned via a mixture of model-free and model-based deep reinforcement learning strategies. We demonstrate, through rigorous numerical and visual experiments, that the learned policy can customize parameters to different settings, and is often more efficient and effective than existing handcrafted criteria. Moreover, we discuss several practical considerations of PnP denoisers, which together with our learned policy yield state-of-the-art results. This advanced performance is prevalent on both linear and nonlinear exemplar inverse imaging problems, and in particular shows promising results on compressed sensing MRI, sparse-view CT, single-photon imaging, and phase retrieval.

arXiv commentThe Journal version (47 pages) of arXiv:2002.09611 (ICML'20 Award Paper); Code is released at https://github.com/Vandermode/TFPnP

References in corpus (41)

• Adam: A Method for Stochastic Optimization
• Continuous control with deep reinforcement learning
• Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
• Playing Atari with Deep Reinforcement Learning
• Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization
• Trust Region Policy Optimization
• Deep Convolutional Neural Network for Inverse Problems in Imaging
• Message Passing Algorithms for Compressed Sensing
• Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)
• Wide-field, high-resolution Fourier ptychographic microscopy
• Non-invasive real-time imaging through scattering layers and around corners via speckle correlations
• Denoising Prior Driven Deep Neural Network for Image Restoration
• Rank Minimization for Snapshot Compressive Imaging
• Tensor-based formulation and nuclear norm regularization for multi-energy computed tomography
• Generalized SURE for Exponential Families: Applications to Regularization
• Deep Unfolding: Model-Based Inspiration of Novel Deep Architectures
• An Online Plug-and-Play Algorithm for Regularized Image Reconstruction
• Learned D-AMP: Principled Neural Network based Compressive Image Recovery
• Unrolled Optimization with Deep Priors
• Bits from Photons: Oversampled Image Acquisition Using Binary Poisson Statistics
• Path-Restore: Learning Network Path Selection for Image Restoration
• Neural Proximal Gradient Descent for Compressive Imaging
• J-MoDL: Joint Model-Based Deep Learning for Optimized Sampling and Reconstruction
• Plug-and-Play Methods Provably Converge with Properly Trained Denoisers
• Learning to Optimize: A Primer and A Benchmark
• Near optimal compressed sensing without priors: Parametric SURE Approximate Message Passing
• DOLPHIn - Dictionary Learning for Phase Retrieval
• On the Effects of Batch and Weight Normalization in Generative Adversarial Networks
• SIMBA: Scalable Inversion in Optical Tomography using Deep Denoising Priors
• Fixed-Point and Objective Convergence of Plug-and-Play Algorithms
• Unified Single-Image and Video Super-Resolution via Denoising Algorithms
• Learning to Paint With Model-based Deep Reinforcement Learning
• Regularized Fourier Ptychography using an Online Plug-and-Play Algorithm
• TorchRadon: Fast Differentiable Routines for Computed Tomography
• On Plug-and-Play Regularization using Linear Denoisers
• Parameterless Optimal Approximate Message Passing
• Phase retrieval from noisy data based on sparse approximation of object phase and amplitude
• Differentiable Linearized ADMM
• Dynamically Unfolding Recurrent Restorer: A Moving Endpoint Control Method for Image Restoration
• Async-RED: A Provably Convergent Asynchronous Block Parallel Stochastic Method using Deep Denoising Priors
• Scalable Plug-and-Play ADMM with Convergence Guarantees

Cited by in corpus (1)

• Deep Plug-and-Play Prior for Hyperspectral Image Restoration