Split-based Feedback Network for Image Super-Resolution

Most existing image super-resolution (SR) methods has achieved superior performance. However, the contrastive learning, which commonly be used in high-level tasks, has not been fully exploited in existing deep learning based image SR methods. In this paper, we propose an image super-resolution network based feedback mechanism to learn abstract representations to explore high information in the representation space. Specifically, we first use the hidden states and constraints in RNN to achieve feedback network. Then, a contrastive learning is used to conduct representation learning by pulling the final SR image to the high resolution image and push the final image to intermediate images. In addition, we introduce a split based feedback block (SPFB) to reduce the redundancy of models for inference acceleration, where the tolerate features with similar patterns but require less computation. Extensive experimental results demonstrate the superiority of the proposed method in comparison with the state-of-the-art methods.