Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining

Cross-Scale Non-Local (CS-NL) Attention
Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining_第1张图片
文中最重要的跨尺度非局部模块就是公式(4),这里内积通过滤波实现,图中的Deconvolution实际上是转置卷积,解卷积和转置卷积是完全不同的概念。
Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining_第2张图片
公式(4)通过如下图理解一目了然,本来可以画个图一清二楚,偏不画。
Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining_第3张图片

你可能感兴趣的:(Paper,reading,深度学习)