Video Enhancement with Task-Oriented Flow (IJCV 2019) Task-Oriented Flow (TOFlow or TOF), a motion representation learned in a selfsupervised, task-specific manner.
EDVR: Video Restoration with Enhanced Deformable Convolutional Networks (CVPR 2019) from SenseTime, CUHK
Project | pyTorch (deperciated) | MMSR
The existence of blur increases the difficulty of motion estimation. EDVR acquire information from multiple frames using alignment, with a slight modification that an image deblurring module is added prior to alignment when there is a blur.
Pyramid, Cascading and Deformable convolution: learn good embedding aligned with next frame
Fuse features of 5 or 7 neighbor frames with Temporal and Spatial Attention
Left: PCD alignment module with Pyramid, Cascading and Deformable convolution; Right: TSA fusion module with Temporal and Spatial Attention
