Revisiting The Evaluation of Class Activation Mapping for Explainability: A Novel Metric and Experimental Analysis

As the request for deep learning solutions increases, the need for\nexplainability is even more fundamental. In this setting, particular attention\nhas been given to visualization techniques, that try to attribute the right\nrelevance to each input pixel with respect to the output of the network. In\nthis paper, we focus on Class Activation Mapping (CAM) approaches, which\nprovide an effective visualization by taking weighted averages of the\nactivation maps. To enhance the evaluation and the reproducibility of such\napproaches, we propose a novel set of metrics to quantify explanation maps,\nwhich show better effectiveness and simplify comparisons between approaches. To\nevaluate the appropriateness of the proposal, we compare different CAM-based\nvisualization methods on the entire ImageNet validation set, fostering proper\ncomparisons and reproducibility.\n