However, they add a residual of the values, passed through a convolution of kernel size 3, which they named Local Interactive Module (LIM).They make the claim in this paper that this scheme outperforms Swin Transformer, and also demonstrate competitive performance against Crossformer....
We propose a NR-IQA model, named STNS-IQA, which combines Swin-Transformer and natural scene statistics. Swin-Transformer is utilized to extract multi-scale information from images. We introduce a feature enhancement module to gather more contextual information. We also incorporate deformable convoluti...