Learning Visual Explanations for DCNN-Based Image Classifiers Using an Attention Mechanism

TL;DR

This paper introduces two novel attention-based methods, L-CAM-Fm and L-CAM-Img, for generating visual explanations of DCNN classifiers, enabling better understanding of model decisions with a single inference pass.

Contribution

The paper proposes two new learning-based XAI methods that integrate an attention mechanism into frozen DCNNs to produce class activation maps for interpretability.

Findings

Achieved competitive results on ImageNet

Provided qualitative insights into classification errors

Enabled explanation generation with a single forward pass

Abstract

In this paper two new learning-based eXplainable AI (XAI) methods for deep convolutional neural network (DCNN) image classifiers, called L-CAM-Fm and L-CAM-Img, are proposed. Both methods use an attention mechanism that is inserted in the original (frozen) DCNN and is trained to derive class activation maps (CAMs) from the last convolutional layer's feature maps. During training, CAMs are applied to the feature maps (L-CAM-Fm) or the input image (L-CAM-Img) forcing the attention mechanism to learn the image regions explaining the DCNN's outcome. Experimental evaluation on ImageNet shows that the proposed methods achieve competitive results while requiring a single forward pass at the inference stage. Moreover, based on the derived explanations a comprehensive qualitative analysis is performed providing valuable insight for understanding the reasons behind classification errors, including possible dataset biases affecting the trained classifier.