Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

About

This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [Erhan et al., 2009], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [Zeiler et al., 2013].

Karen Simonyan, Andrea Vedaldi, Andrew Zisserman · 2013

Related benchmarks

Task	Dataset	Result
Image Classification	MNIST (test)	Accuracy91.35	894
Image Classification	SVHN (test)	Accuracy63.74	470
Explainability	ImageNet (val)	Insertion36.3	104
Faithfulness Evaluation	TellMeWhy	AUC π-Soft-NS0.309	67
Faithfulness Evaluation	WikiBio	AUC π-Soft-NS0.345	67
Feature Relevance Evaluation	ImageNet (test)	R (Feature Relevance)0.4	60
XAI Fidelity Evaluation	ImageNet vehicles	Fidelity Score72.5	58
Object Localization	ImageNet vehicles	IoU60.6	58
Attribution Fidelity	ImageNet 1,000 images (val)	µFidelity0.192	48
Deletion	ImageNet 2,000 images (val)	Deletion Score0.174	48