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

Wed, 25 Feb 2026 00:00:00 +0000

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.

Apprentissage auto-supervisé pour la reconstruction de phase

Fri, 01 Aug 2025 00:00:00 +0000

Résumé: Ces dernières années, les réseaux de neurones profonds sont apparus comme une solution aux problèmes d’imagerie inverse. Ces réseaux sont généralement entraînés à partir de paires d’images : une image dégradée et une image de haute qualité, appelée ‘vérité terrain’. Cependant, en imagerie médicale et scientifique, le manque de données complètement échantillonnées limite l’apprentissage supervisé. Des avancées récentes permettent de reconstruire des images à partir des seules mesures, éliminant le besoin de références. Ces méthodes restent toutefois limitées aux problèmes linéaires, excluant les problèmes non linéaires comme la reconstruction de phase. Nous proposons une méthode auto-supervisée qui surmonte cette limitation dans le cas de la reconstruction de phase en exploitant l’invariance naturelle des images aux translations.

Abstract: In recent years, deep neural networks have emerged as a solution for inverse imaging problems. These networks are generally trained using pairs of images: one degraded and the other of high quality, the latter being called ‘ground truth’. However, in medical and scientific imaging, the lack of fully sampled data limits supervised learning. Recent advances have made it possible to reconstruct images from measurement data alone, eliminating the need for references. However, these methods remain limited to linear problems, excluding non-linear problems such as phase retrieval. We propose a self-supervised method that overcomes this limitation in the case of phase retrieval by using the natural invariance of images to translations.

V. Sechaud | Laurent Jacques

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

Apprentissage auto-supervisé pour la reconstruction de phase