Viet-Anh on Software Logo

What is: Cross-Scale Non-Local Attention?

SourceImage Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining
Year2000
Data SourceCC BY-SA - https://paperswithcode.com

Cross-Scale Non-Local Attention, or CS-NL, is a non-local attention module for image super-resolution deep networks. It learns to mine long-range dependencies between LR features to larger-scale HR patches within the same feature map. Specifically, suppose we are conducting an s-scale super-resolution with the module, given a feature map XX of spatial size (W,H)(W, H), we first bilinearly downsample it to YY with scale ss, and match the p×pp\times p patches in XX with the downsampled p×pp \times p candidates in YY to obtain the softmax matching score. Finally, we conduct deconvolution.on the score by weighted adding the patches of size (sp,sp)\left(sp, sp\right) extracted from XX. The obtained ZZ of size (sW,sH)(sW, sH) will be ss times super-resolved than XX.