Sharpen Focus: Learning with Attention Separability and Consistency

TL;DR

This paper introduces a new learning framework that enhances class-discriminative attention in CNNs by enforcing attention separability and consistency, leading to improved classification accuracy across multiple benchmarks.

Contribution

The paper proposes novel objectives for attention separability and cross-layer consistency, making discriminative attention an integral part of the training process.

Findings

Improved classification accuracy on CIFAR-100 (+3.33%)

Enhanced attention discriminability reducing visual confusion

Consistent performance gains across multiple datasets

Abstract

Recent developments in gradient-based attention modeling have seen attention maps emerge as a powerful tool for interpreting convolutional neural networks. Despite good localization for an individual class of interest, these techniques produce attention maps with substantially overlapping responses among different classes, leading to the problem of visual confusion and the need for discriminative attention. In this paper, we address this problem by means of a new framework that makes class-discriminative attention a principled part of the learning process. Our key innovations include new learning objectives for attention separability and cross-layer consistency, which result in improved attention discriminability and reduced visual confusion. Extensive experiments on image classification benchmarks show the effectiveness of our approach in terms of improved classification accuracy, including CIFAR-100 (+3.33%), Caltech-256 (+1.64%), ILSVRC2012 (+0.92%), CUB-200-2011 (+4.8%) and PASCAL VOC2012 (+5.73%).