Translation-based Video-to-Video Synthesis

arXiv:2404.04283v1 Announce Type: cross
Abstract: Translation-based Video Synthesis (TVS) has emerged as a vital research area in computer vision, aiming to facilitate the transformation of videos between distinct domains while preserving both temporal continuity and underlying content features. This technique has found wide-ranging applications, encompassing video super-resolution, colorization, segmentation, and more, by extending the capabilities of traditional image-to-image translation to the temporal domain. One of the principal challenges faced in TVS is the inherent risk of introducing flickering artifacts and …

abstract applications arxiv capabilities colorization computer computer vision continuity cs.cv cs.lg domains eess.iv features found research resolution segmentation synthesis temporal transformation translation type video videos video-to-video vision vital