Looking beyond input frames: Self-supervised adaptation for video super-resolution

Jinsu Yoo, Jihoon Nam, Sungyong Baik, Tae Hyun Kim

Abstract: Highlights•Improving restoration performances of pre-trained VSR networks by leveraging test LR video frames.•Allowing large scale factors for test-time adaptive SR.•Showing flexibility in ways to distill knowledge.•Removing undesirable initial artifacts across video frames.•Consistently elevating performances on various VSR networks and datasets.