Learnable Descent Algorithm for Nonsmooth Nonconvex Image Reconstruction
arXiv:2007.11245
Abstract
We propose a general learning based framework for solving nonsmooth and nonconvex image reconstruction problems. We model the regularization function as the composition of the norm and a smooth but nonconvex feature mapping parametrized as a deep convolutional neural network. We develop a provably convergent descent-type algorithm to solve the nonsmooth nonconvex minimization problem by leveraging the Nesterov's smoothing technique and the idea of residual learning, and learn the network parameters such that the outputs of the algorithm match the references in training data. Our method is versatile as one can employ various modern network structures into the regularization, and the resulting network inherits the guaranteed convergence of the algorithm. We also show that the proposed network is parameter-efficient and its performance compares favorably to the state-of-the-art methods in a variety of image reconstruction problems in practice.
References in corpus (13)
- Adam: A Method for Stochastic Optimization
- Wide Residual Networks
- A guide to convolution arithmetic for deep learning
- AMP-Inspired Deep Networks for Sparse Linear Inverse Problems
- DR2-Net: Deep Residual Reconstruction Network for Image Compressive Sensing
- A General Iterative Shrinkage and Thresholding Algorithm for Non-convex Regularized Optimization Problems
- Splitting methods with variable metric for KL functions
- Theoretical Linear Convergence of Unfolded ISTA and its Practical Weights and Thresholds
- Multi-Channel Deep Networks for Block-Based Image Compressive Sensing
- Maximal Sparsity with Deep Networks?
- Differentiable Linearized ADMM
- Proximal alternating minimization and projection methods for nonconvex problems. An approach based on the Kurdyka-Lojasiewicz inequality
- Minimization of nonsmooth nonconvex functions using inexact evaluations and its worst-case complexity