Recurrent Feature Updating Network for Video Super-resolution

Temporal modeling is the essential to achieve video super-resolution. Most models use alignment or recurrent methods directly exploit temporal information of consecutive frames. However, feature extracted from input frames coarse, which affects performance and generalization ability model. Thus, in this paper, we propose investigate role updating networks. The update module proposed optimize improve accuracy features. We design model as a multi-stage form generalization. Moreover, further model, difference supplement module. It allows differences between groups complement missing output each stage. Experiments demonstrate that our achieves 27.78 dB, 30.44 39.38 dB on Vid4, SPMCS, UDM10, respectively, indicates state-of-the-art Code available at https://github.com/gfli123/RFUN.