Grad-CAM++: Generalized Gradient-Based Visual Explanations for Deep Convolutional Networks

By Aditya Chattopadhyay, Anirban Sarkar, Prantik Howlader, Vineeth Nallure Balasubramanian. 2017

3,283 citations recorded by the source corpus.

Over the last decade, Convolutional Neural Network (CNN) models have been highly successful in solving complex vision based problems. However, deep models are perceived as "black box" methods considering the lack of understanding of their internal functioning. There has been a significant recent interest to develop explainable deep learning models, and this paper is an effort in this direction. Building on a recently proposed method called Grad-CAM, we propose Grad-CAM++ to provide better visual explanations of CNN model predictions (when compared to Grad-CAM), in terms of better localization of objects as well as explaining occurrences of multiple objects of a class in a single image. We provide a mathematical explanation for the proposed method, Grad-CAM++, which uses a weighted combination of the positive partial derivatives of the last convolutional layer feature maps with respect to a specific class score as weights to generate a visual explanation for the class label under consideration. Our extensive experiments and evaluations, both subjective and objective, on standard datasets showed that Grad-CAM++ indeed provides better visual explanations for a given CNN architecture when compared to Grad-CAM.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety, Robustness & Interpretability

Citations in the corpus

SentiNet: Detecting Localized Universal Attacks Against Deep Learning Systems (2018, methodology): “A general guideline towards selecting k is to set it to be slightly A particularly suitable approach for our goal is Grad-CAM [6], a model-interpretation technique that identiﬁes contiguous spatial regions of an input without requiring modi-ﬁcations to the original model.”
Interpretable and Fine-Grained Visual Explanations for Convolutional Neural Networks (2019, methodology): “Prominent methods of this category are CAM (Class Activation Mapping) [53] and its generalizations Grad-CAM [35] and Grad-CAM++ [7].”
A case-based interpretable deep learning model for classification of mass lesions in digital mammography (2021, methodology): “44) and GradCAM++ (ref.”
Remembering for the Right Reasons: Explanations Reduce Catastrophic Forgetting (2020, methodology): “Activation-based methods (Zhou et al., 2016; Selvaraju et al., 2017; Chattopadhay et al., 2018) use a weighted linear combination of feature maps whereas gradient-based methods (Baehrens et al., 2010; Sundararajan et al., 2017a; Springenberg et al., 2014; Shrikumar et al., 2017; Zhang et al., 2018)…”
Aggregating Nested Transformers (2021, result): “Figure A4 compares the qualitative results of CAM, including Grad-CAM++ [5] with ResNet50 [23], DeiT with Rollout attention [1], and our NesT CAM [65].”
Explaining Knowledge Distillation by Quantifying the Knowledge (2020, methodology): “Some methods explored the saliency of the input image using intermediate-layer features, such as CAM [52], Grad-CAM [34], and Grad-CAM++ [2].”
Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images (2021, neutral): “There are many popular visualization techniques such as SmoothGrad [33], Grad-CAM [34], Grad-CAM++ [35], Score-CAM [26].”
SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (2023, methodology): “…pixel-space sensitivity map methods (Simonyan et al., 2013; Zeiler & Fergus, 2014; Springenberg et al., 2014; Smilkov et al., 2017; Sundararajan et al., 2017) and class-discriminative localization methods (Zhou et al., 2016; Selvaraju et al., 2017; Chattopadhay et al., 2018; Petsiuk et al., 2018).”
Improving Plasticity in Online Continual Learning via Collaborative Learning (2023, methodology): “5, we use GradCAM++ [8] on the training set of ImageNet-100 (M=5k) at the end of the training of ER and GSA.”
Proactive Pseudo-Intervention: Causally Informed Contrastive Learning For Interpretable Vision Models (2020, background): “Grad-CAM++ [7] proposes to introduce higher-order derivatives to capture pixel-level importance, while its high computational cost in calculating the second-and third-order derivatives makes it impractical for training purposes.”
Automatic Classification of Defective Photovoltaic Module Cells in Electroluminescence Images (2018, methodology): “The solar cells are additionally overlaid by CAMs determined using Grad-CAM++ [61].”
Deep Learning for Medical Anomaly Detection – A Survey (2020, background): “Grad-Cam however does not handle instances where multiple occurrences of the same object are in the input image, and in such instances it fails to properly localise the multiple instances [106].”
Saliency Guided Experience Packing for Replay in Continual Learning (2021, methodology): “To assess the sensitivity of EPR to the choice of saliency algorithms, we also used Grad-CAM++ [6], Smooth-Grad [46] and FullGrad [47] as saliency method in EPR.”
Transformers in Remote Sensing: A Survey (2022, methodology): “using Grad-CAM++ [34], of different ImageNet (IMP) and remote sensing pre-trained (RSP) models.”
Ablation Studies in Artificial Neural Networks (2019, background): “Some recent work on this matter demonstrated how the contribution of a network’s input elements to its decision could be explained by means of Deep Taylor Decomposition [36] or Gradient-weighted Class Activation Mapping (Grad-CAM) [37,38].”