EDVR: Video Restoration With Enhanced Deformable Convolutional Networks

TLDR

Video restoration tasks such as super‑resolution and deblurring are gaining attention, and the REDS benchmark in NTIRE19 challenges methods on large‑motion alignment and diverse frame fusion. In this work, we propose a novel Video Restoration framework with Enhanced Deformable convolutions, termed EDVR, to address these challenges. EDVR uses a Pyramid, Cascading and Deformable (PCD) alignment module for large‑motion feature alignment and a Temporal and Spatial Attention (TSA) fusion module to emphasize important features, with code released at GitHub. EDVR won the NTIRE19 video restoration and enhancement challenges, outperforming the runner‑up by a large margin, and achieved superior performance over state‑of‑the‑art methods on video super‑resolution and deblurring.

Abstract

Video restoration tasks, including super-resolution, deblurring, etc, are drawing increasing attention in the computer vision community. A challenging benchmark named REDS is released in the NTIRE19 Challenge. This new benchmark challenges existing methods from two aspects: (1) how to align multiple frames given large motions, and (2) how to effectively fuse different frames with diverse motion and blur. In this work, we propose a novel Video Restoration framework with Enhanced Deformable convolutions, termed EDVR, to address these challenges. First, to handle large motions, we devise a Pyramid, Cascading and Deformable (PCD) alignment module, in which frame alignment is done at the feature level using deformable convolutions in a coarse-to-fine manner. Second, we propose a Temporal and Spatial Attention (TSA) fusion module, in which attention is applied both temporally and spatially, so as to emphasize important features for subsequent restoration. Thanks to these modules, our EDVR wins the champions and outperforms the second place by a large margin in all four tracks in the NTIRE19 video restoration and enhancement challenges. EDVR also demonstrates superior performance to state-of-the-art published methods on video super-resolution and deblurring. The code is available at https://github.com/xinntao/EDVR.