Invertible generative models for inverse problems: mitigating representation error and dataset bias

Abstract

Trained generative models have shown remarkable performance as priors for inverse problems in imaging -- for example, Generative Adversarial Network priors permit recovery of test images from 5-10x fewer measurements than sparsity priors. Unfortunately, these models may be unable to represent any particular image because of architectural choices, mode collapse, and bias in the training dataset. In this paper, we demonstrate that invertible neural networks, which have zero representation error by design, can be effective natural signal priors at inverse problems such as denoising, compressive sensing, and inpainting. Given a trained generative model, we study the empirical risk formulation of the desired inverse problem under a regularization that promotes high likelihood images, either directly by penalization or algorithmically by initialization. For compressive sensing, invertible priors can yield higher accuracy than sparsity priors across almost all undersampling ratios, and due to their lack of representation error, invertible priors can yield better reconstructions than GAN priors for images that have rare features of variation within the biased training set, including out-of-distribution natural images. We additionally compare performance for compressive sensing to unlearned methods, such as the deep decoder, and we establish theoretical bounds on expected recovery error in the case of a linear invertible model.

Camera ready version for ICML 2020, paper 2655

References in corpus (9)

• Auto-Encoding Variational Bayes
• Density estimation using Real NVP
• Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks
• Analyzing Inverse Problems with Invertible Neural Networks
• Statistical Compressed Sensing of Gaussian Mixture Models
• i-RevNet: Deep Invertible Networks
• Finger-GAN: Generating Realistic Fingerprint Images Using Connectivity Imposed GAN
• Invert to Learn to Invert
• Latent Convolutional Models

Cited by in corpus (6)

• Robust Compressed Sensing MRI with Deep Generative Priors
• Quantum Algorithms for the Pathwise Lasso
• Subsampled Fourier Ptychography using Pretrained Invertible and Untrained Network Priors
• D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
• Class-Specific Blind Deconvolutional Phase Retrieval Under a Generative Prior
• Reducing the Representation Error of GAN Image Priors Using the Deep Decoder