Abstract

Video Super Resolution (VSR) has emerged as a crucial task in the field of Computer Vision due to its diverse applications. In this paper, we propose the Recurrent Back-Projection Generative Adversarial Network (RBPGAN) for VSR, aiming to generate temporally coherent videos while preserving spatial details. RBPGAN integrates two state-of-the-art models to leverage their strengths without compromising the accuracy of the output video. The generator in our model is inspired by the RBPN system, while the discriminator draws from TecoGAN. Additionally, we employ a Ping-Pong loss to enhance temporal consistency over time. Our approach results in a model that surpasses previous works in producing temporally consistent details, as demonstrated through both qualitative and quantitative evaluations across different datasets.

Statistics