Abstract:
We present a self-supervised and self-calibrating multi-shot approach to imaging through atmospheric turbulence, called TurbuGAN. Our approach requires no paired training...View moreMetadata
Abstract:
We present a self-supervised and self-calibrating multi-shot approach to imaging through atmospheric turbulence, called TurbuGAN. Our approach requires no paired training data, adapts itself to the distribution of the turbulence, flexibly leverages prior knowledge of images from different domains, outperforms existing approaches, and can generalize from tens to tens of thousands of measurements. We achieve such functionality through an adversarial sensing framework adapted from CryoGAN [1], with a discriminator network to match the distributions of captured and simulated measurements. We extend CryoGAN by (1) generalizing the forward measurement model to incorporate physically accurate and computationally efficient models for light propagation through anisoplanatic turbulence, (2) enabling adaptation to slightly misspecified forward models, and (3) flexibly leveraging prior knowledge of images from different domains using pretrained generative networks. We validate the effectiveness of TurbuGAN with simulated experiments using realistic models for atmospheric turbulence-induced distortion. An extended version of this manuscript is available [2].
Date of Conference: 31 October 2022 - 02 November 2022
Date Added to IEEE Xplore: 07 March 2023
ISBN Information: