The core message of this paper is that introducing self-supervised constraints can significantly improve the performance of existing image super-resolution models. The authors propose a novel self-supervised constraint framework, SSC-SR, which leverages data augmentation and a dual asymmetric architecture to refine and stabilize super-resolution techniques.
The proposed Learnable Collaborative Attention (LCoA) mechanism encodes inductive biases of learnable sparsity and weight sharing into non-local modeling, significantly improving the computational efficiency of single image super-resolution without compromising reconstruction quality.
The proposed multi-scale attention network (MAN) couples classical multi-scale mechanism with large kernel attention to effectively capture global and local information for single image super-resolution, achieving state-of-the-art performance with varied trade-offs between model complexity and computations.
A novel super-resolution reconstruction algorithm that achieves significant accuracy improvement through a unique design while maintaining low computational complexity.