Semi-Supervised Learning for Low-light Image Restoration through Quality Assisted Pseudo-Labeling

Convolutional neural networks have been successful in restoring images captured under poor illumination conditions. Nevertheless, such approaches require a large number of paired low-light and ground truth images for training. Thus, we study the problem of semi-supervised learning for low-light image restoration when limited low-light images have ground truth labels. Our main contributions in this work are twofold. We first deploy an ensemble of low-light restoration networks to restore the unlabeled images and generate a set of potential pseudo-labels. We model the contrast distortions in the labeled set to generate different sets of training data and create the ensemble of networks. We then design a contrastive self-supervised learning based image quality measure to obtain the pseudo-label among the images restored by the ensemble. We show that training the restoration network with the pseudo-labels allows us to achieve excellent restoration performance even with very few labeled pairs. We conduct extensive experiments on three popular low-light image restoration datasets to show the superior performance of our semi-supervised low-light image restoration compared to other approaches. Project page is available at https://github.com/sameerIISc/SSL-LLR.

1. Introduction

Restoring low-light images is challenging because of the presence of multiple distortions such as poor contrast, noise and color cast. While several traditional low-light image restoration (LLIR) methods exist, convolutional neural networks (CNNs) have tremendously succeeded in such a complex restoration task. However, CNNs suffer from the shortcoming of requiring large amounts of aligned low-light and clean well-exposed image pairs for training. Moreover, such models are camera distortion specific and require data to be collected everytime a new model needs to be trained for a different camera.