A Non-Local Block With Adaptive Regularization Strategy

Non-local block (NLB) is a breakthrough technology in computer vision. It greatly boosts the capability of deep convolutional neural networks (CNNs) to capture long-range dependencies. As the critical component of NLB, non-local operation can be considered a network-based implementation of the well-known non-local means filter (NLM). Drawing on the solid theoretical foundation of NLM, we provide an innovative interpretation of the non-local operation. Specifically, it is formulated as an optimization problem regularized by Shannon entropy with a fixed parameter. Building on this insight, we further introduce an adaptive regularization strategy to enhance NLB and get a novel non-local block named ARNLB. Preliminary experiments on semantic segmentation demonstrate its effectiveness. The code of ARNLB is accessible at http://www.diplab.net/lunwen/arnlb.htm