Learning to reconstruct from saturated data: audio declipping and high-dynamic range imaging

TL;DR

This paper introduces a self-supervised learning approach for reconstructing audio and images from saturated, clipped data, enabling effective recovery without ground truth references, and extending methods to non-linear inverse problems.

Contribution

It extends self-supervised learning to non-linear inverse problems like declipping and HDR imaging, providing conditions and a loss function for training from saturated data alone.

Findings

Approaches perform nearly as well as fully supervised methods.

Effective reconstruction achieved using only clipped measurements for training.

Applicable to both audio and image data.

Abstract

Learning based methods are now ubiquitous for solving inverse problems, but their deployment in real-world applications is often hindered by the lack of ground truth references for training. Recent self-supervised learning strategies offer a promising alternative, avoiding the need for ground truth. However, most existing methods are limited to linear inverse problems. This work extends self-supervised learning to the non-linear problem of recovering audio and images from clipped measurements, by assuming that the signal distribution is approximately invariant to changes in amplitude. We provide sufficient conditions for learning to reconstruct from saturated signals alone and a self-supervised loss that can be used to train reconstruction networks. Experiments on both audio and image data show that the proposed approach is almost as effective as fully supervised approaches, despite relying solely on clipped measurements for training.