Grad-CAM++ is Equivalent to Grad-CAM With Positive Gradients

TL;DR

This paper demonstrates that Grad-CAM++ is essentially equivalent to a simplified version of Grad-CAM that uses only positive gradients, clarifying the relationship between these popular visualization methods.

Contribution

The paper reveals the theoretical equivalence between Grad-CAM++ and a positive-gradient-based variant of Grad-CAM, simplifying understanding of their differences.

Findings

Grad-CAM++ is equivalent to Grad-CAM with positive gradients

Positive gradients are sufficient for effective visual explanations

Simplifies the interpretation of Grad-CAM++ results

Abstract

The Grad-CAM algorithm provides a way to identify what parts of an image contribute most to the output of a classifier deep network. The algorithm is simple and widely used for localization of objects in an image, although some researchers have point out its limitations, and proposed various alternatives. One of them is Grad-CAM++, that according to its authors can provide better visual explanations for network predictions, and does a better job at locating objects even for occurrences of multiple object instances in a single image. Here we show that Grad-CAM++ is practically equivalent to a very simple variation of Grad-CAM in which gradients are replaced with positive gradients.