DCAN: Deep Consecutive Attention Network for Video Super Resolution

Slow motion is visually attractive in video applications and gets more attention in video super-resolution (VSR). To generate the high-resolution (HR) center frame with its neighbor HR frames from the low-resolution (LR) of two frames. Two sub-tasks are required, including video super-resolution (VSR) and video frame interpolation (VFI). However, the interpolation approach does not successfully extract low-level features to achieve the acceptable result of space-time video super-resolution. Therefore, the restoration performance of existing systems is constrained due to rarely considering the spatial-temporal correlation and the long-term temporal context concurrently. To this extent, we propose a deep consecutive attention network-based method to generate attentive features to get HR slow-motion frames. A channel attention module and an attentive temporal feature module are designed to improve the perceptual quality of predicted interpolation feature frames. The experimental results show the proposed method outperforms 0.17 dB in an average PSNR compared to the state-of-the-art baseline method.