{ "id": "epj2bn1d6v", "slug": "grad-cam", "title": "Grad-CAM", "updated": "2026-10-15", "topic_path": [ { "id": "technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/topics/technology" }, { "id": "technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing" }, { "id": "technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai" }, { "id": "technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.ml_computation" }, { "id": "technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep" }, { "id": "technology.computing.ai.neural_deep.alpha-a-to-h", "label": "Titles A to H", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep.alpha-a-to-h" } ], "geo": [ { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Non-geographic · 2001 to 2020: Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep", "path": [ { "id": "geo.nongeo", "label": "Non-geographic", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo" }, { "id": "geo.nongeo.t2001", "label": "Non-geographic · 2001 to 2020", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001" }, { "id": "geo.nongeo.t2001.technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology" }, { "id": "geo.nongeo.t2001.technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing" }, { "id": "geo.nongeo.t2001.technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai" }, { "id": "geo.nongeo.t2001.technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.ml_computation" }, { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep" } ] } ], "excerpt": "Grad-CAM (Gradient-weighted Class Activation Mapping) is a visualization technique producing heatmaps of the image regions a convolutional neural network relied on for a prediction, introduced in 2016.", "snippet": "Grad-CAM (Gradient-weighted Class Activation Mapping) is a visualization technique producing heatmaps of the image regions a convolutional neural network relied on for a prediction, introduced in 2016.", "node": "technology.computing.ai.neural_deep.alpha-a-to-h", "markdown": "# Grad-CAM\n\nGrad-CAM (Gradient-weighted Class Activation Mapping) is a visualization technique that produces a class-specific saliency heatmap showing which image regions a convolutional neural network relied on for a particular prediction, by combining backpropagated gradients with the feature maps of a convolutional layer.[1](https://arxiv.org/abs/1610.02391) It requires only one forward pass and a partial backward pass per image and works on unmodified networks.[1](https://arxiv.org/abs/1610.02391)\n\n| Key fact | Detail |\n|---|---|\n| Output | A coarse, class-discriminative heatmap the size of the chosen layer's feature maps (14×14 for the last convolutional layers of VGG and AlexNet), upsampled and overlaid on the image[1](https://arxiv.org/abs/1610.02391) |\n| Core computation | \\( L^{c} = \\mathrm{ReLU}(\\sum_{k} \\alpha_{k}^{c} \\cdot A^{k}) \\), with \\( \\alpha_{k}^{c} \\) from global-average-pooled gradients of the pre-softmax class score[1](https://arxiv.org/abs/1610.02391) |\n| Cost | One forward plus partial backward pass per image, typically an order of magnitude cheaper than occlusion-based perturbation methods[1](https://arxiv.org/abs/1610.02391) |\n| Typical layer | The final convolutional layer, which balances spatial information against semantic relevance[2](https://www.nature.com/articles/s41598-025-25839-y) |\n| Faithfulness caveat | Because gradients are averaged, Grad-CAM is not guaranteed to highlight only the locations the model used and can produce misleading explanations[3](https://doi.org/10.48550/arxiv.2011.08891) |\n| Main tooling | The PyTorch package pytorch-grad-cam implements Grad-CAM and many variants with batch support and smoothing options[4](https://github.com/jacobgil/pytorch-grad-cam) |\n| Adoption | Published at ICCV 2017 (Venice, 22–29 October 2017) by Selvaraju, Cogswell, Das, Vedantam, Parikh, and Batra[5](https://ai.meta.com/research/publications/grad-cam-visual-explanations-from-deep-networks-via-gradient-based-localization/) |\n\n## How it works\n\nFor a class \\( c \\), Grad-CAM takes the gradient of the class score \\( y^{c} \\) taken before the softmax, with respect to the activation \\( A^{k} \\) of each feature map \\( k \\) in a chosen convolutional layer. These gradients are global-average-pooled over the spatial dimensions to give one importance weight per channel,\n\n\\[ \\alpha_{k}^{c} = \\frac{1}{Z} \\sum_{i} \\sum_{j} \\frac{\\partial y^{c}}{\\partial A_{ij}^{k}} \\]\n\nand the heatmap is the ReLU-clipped weighted sum of the feature maps,[1](https://arxiv.org/abs/1610.02391) • [2](https://www.nature.com/articles/s41598-025-25839-y)\n\n\\[ L^{c}_{\\text{Grad-CAM}} = \\mathrm{ReLU}\\left( \\sum_{k} \\alpha_{k}^{c} \\cdot A^{k} \\right) \\]\n\nThe ReLU keeps only features with a positive influence on the class of interest; removing it increases top-1 localization error by 15.3% on ILSVRC-15 val.[6](https://link.springer.com/article/10.1007/s11263-019-01228-7) Global average pooling of the gradients works empirically better than global max pooling, which lowers localization ability.[1](https://arxiv.org/abs/1610.02391) • [6](https://link.springer.com/article/10.1007/s11263-019-01228-7)\n\nGrad-CAM is a strict generalization of Class Activation Mapping (CAM). CAM computes the class score \\( S_{c} = \\sum_{k} w_{k}^{c} \\cdot F_{k} \\) using the learned output-layer weights, where \\( F_{k} \\) is the global-average-pooled activation of unit \\( k \\); the corresponding spatial CAM map is the weighted sum of the unpooled feature maps with the same weights.[7](https://openaccess.thecvf.com/content_cvpr_2016/papers/Zhou_Learning_Deep_Features_CVPR_2016_paper.pdf) Up to a proportionality constant normalized out during visualization, CAM's \\( w_{k}^{c} \\) equals Grad-CAM's \\( \\alpha_{k}^{c} \\), so for fully convolutional architectures CAM is a special case of Grad-CAM.[1](https://arxiv.org/abs/1610.02391)\n\n## How it is done\n\nA practitioner runs these steps, as implemented in reference PyTorch code:[8](https://github.com/kazuto1011/grad-cam-pytorch/blob/master/grad_cam.py)\n\n1. Run a forward pass on the preprocessed image and record the activations of the target convolutional layer, usually via forward hooks.\n2. Backpropagate the score \\( y^{c} \\) for the class of interest (before softmax) and capture the gradients at the same layer.\n3. Pool the gradients spatially with an operation such as `F.adaptive_avg_pool2d` to obtain the channel weights \\( \\alpha_{k}^{c} \\).\n4. Multiply each feature map by its weight, sum over channels, and apply ReLU.\n5. Bilinearly upsample the heatmap to the input image shape and min-max normalize it, then overlay it on the original image.\n\nThe target layer is usually the last convolutional layer, which captures the most relevant high-level features while retaining some spatial information; earlier layers or averaging across layers can give more accurate explanations for fine details.[2](https://www.nature.com/articles/s41598-025-25839-y) Practical layer choices documented in the pytorch-grad-cam package include `model.layer4[-1]` for ResNet18/50 and `model.blocks[-1].norm1` for ViT; a `reshape_transform` argument adapts the method to non-CNN architectures by converting activations back into a multi-channel image, for example by removing the class token in a vision transformer.[4](https://github.com/jacobgil/pytorch-grad-cam)\n\n## Origin\n\nGrad-CAM was reported by Ramprasaath R. Selvaraju and colleagues in \"Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization\", posted to arXiv in 2016 and published at ICCV 2017 in Venice.[9](https://doi.org/10.48550/arxiv.1610.02391) • [10](https://ieeexplore.ieee.org/document/8237336)\n\nIt built on earlier work the authors cite as precursors: Zeiler and Fergus's \"Visualizing and Understanding Convolutional Networks\" (2013), which the reference implementation credits for occlusion sensitivity,[8](https://github.com/kazuto1011/grad-cam-pytorch/blob/master/grad_cam.py) Springenberg and colleagues' \"Striving for Simplicity: The All Convolutional Net\" (2014), which introduced guided backpropagation,[11](https://doi.org/10.48550/arxiv.1412.6806) and Zhou, Khosla, Lapedriza, Oliva, and Torralba's \"Learning Deep Features for Discriminative Localization\" (2015 arXiv, CVPR 2016), which introduced CAM.[12](https://doi.org/10.48550/arxiv.1512.04150) CAM requires feature maps to feed directly through global average pooling into the softmax, so fully connected layers must be replaced with convolutional ones and the network re-trained, trading model performance for transparency.[7](https://openaccess.thecvf.com/content_cvpr_2016/papers/Zhou_Learning_Deep_Features_CVPR_2016_paper.pdf) Grad-CAM removes this restriction: it applies to CNNs with fully connected layers, structured-output models such as captioning, and multi-modal tasks such as visual question answering, with no architectural changes or re-training.[1](https://arxiv.org/abs/1610.02391)\n\n## Variants\n\n**Guided Grad-CAM** fuses the Grad-CAM heatmap with guided backpropagation by element-wise multiplication after bilinear upsampling, yielding a high-resolution visualization that remains class-discriminative.[1](https://arxiv.org/abs/1610.02391)\n\n**Grad-CAM++**, introduced by Chattopadhyay, Sarkar, Howlader, and Balasubramanian (2017), changes the weighting rule to \\( a_{k}^{c} = \\sum_{i} \\sum_{j} w_{ij}^{kc}\\, \\mathrm{ReLU}(\\partial y^{c}/\\partial A_{ij}^{k}) \\), a weighted combination of positive partial derivatives that emphasizes strongly contributing pixels and improves explanation of images containing multiple objects or instances.[13](https://doi.org/10.48550/arxiv.1710.11063) • [2](https://www.nature.com/articles/s41598-025-25839-y)\n\n**Score-CAM**, introduced by Wang and colleagues (2019), removes the dependence on backpropagation entirely: activation maps are used as masks, the masked images are forwarded through the network, and the target-class confidence serves as the channel weight. This avoids gradient artifacts but is expensive, since a separate forward pass may be needed per activation map.[14](https://doi.org/10.48550/arxiv.1910.01279)\n\n**HiResCAM**, introduced by Draelos and Carin (2020), applies the gradients element-wise instead of averaging them, and is provably guaranteed to highlight only the locations used by any CNN ending in one fully connected layer.[3](https://doi.org/10.48550/arxiv.2011.08891)\n\nFurther variants modify the weighting signal or the scoring target. Axiom-based Grad-CAM (Fu and colleagues, 2020) reformulates the weighting to satisfy stated axioms,[15](https://doi.org/10.48550/arxiv.2008.02312) and Integrated Grad-CAM (Sattarzadeh and colleagues, 2021) scores channels with [integrated gradients](https://www.edgepedia.org/integrated-gradients).[16](https://doi.org/10.48550/arxiv.2102.07805) LayerCAM weights activations by positive gradients at individual spatial locations, so it can produce reliable maps from different CNN layers.[17](https://arxiv.org/pdf/2608.12299.pdf) LibraGrad identifies unbalanced gradient flow during backpropagation as the root cause of unfaithful gradient-based attributions and corrects it by pruning and scaling backward paths without touching the forward pass; it improves gradient-based methods including Grad-CAM++, HiResCAM, and FullGrad, and works even on the attention-free MLP-Mixer.[18](https://openaccess.thecvf.com/content/CVPR2025/papers/Mehri_LibraGrad_Balancing_Gradient_Flow_for_Universally_Better_Vision_Transformer_Attributions_CVPR_2025_paper.pdf) • [19](https://doi.org/10.48550/arxiv.2411.16760)\n\n## Applications\n\n**Weakly supervised localization and segmentation.** Replacing CAM maps with Grad-CAM maps in a weakly supervised segmentation pipeline raises Intersection over Union on [PASCAL VOC](https://www.edgepedia.org/pascal-voc) 2012 from 44.6 to 49.6, and Grad-CAM outperformed previous methods on the ILSVRC-15 weakly supervised localization task.[1](https://arxiv.org/abs/1610.02391) Grad-CAM also beats HiResCAM at weakly supervised segmentation, hypothesized to result from Grad-CAM's tendency to expand attention beyond the regions the model actually used.[3](https://doi.org/10.48550/arxiv.2011.08891)\n\n**Model debugging and human-facing explanation.** Human studies showed that Grad-CAM helps untrained users discern a stronger network from a weaker one even when both make identical predictions.[5](https://ai.meta.com/research/publications/grad-cam-visual-explanations-from-deep-networks-via-gradient-based-localization/)\n\n**Sensitive domains.** CAM and Grad-CAM have been deployed in sensitive applications including medical imaging; Grad-CAM is often chosen because its explanations can be computed for any CNN architecture, while CAM is restricted to a subset of architectures.[3](https://doi.org/10.48550/arxiv.2011.08891)\n\n**Transformers.** With a reshape transform, CAM methods can be applied to vision transformers and similar architectures; Attention Guided CAM (2024) aggregates gradients propagated to each self-attention layer guided by sigmoid-normalized self-attention scores, reaching 73.41% pixel accuracy and 52.12% IoU on ImageNet weakly supervised localization with ViT-base.[4](https://github.com/jacobgil/pytorch-grad-cam) • [20](https://arxiv.org/abs/2402.04563)\n\n## Limitations and alternatives\n\n**Resolution.** The raw heatmap is the size of the convolutional feature maps, 14×14 for the last convolutional layers of VGG and AlexNet, so all fine structure comes from upsampling.[1](https://arxiv.org/abs/1610.02391)\n\n**Layer choice.** Maps become progressively worse at earlier convolutional layers, which have smaller receptive fields and focus on less semantic local features.[1](https://arxiv.org/abs/1610.02391) • [6](https://link.springer.com/article/10.1007/s11263-019-01228-7)\n\n**Faithfulness.** Because of the gradient averaging step, Grad-CAM sometimes highlights locations the model did not actually use and is not guaranteed to reflect the locations used for prediction, so it can produce misleading explanations.[3](https://doi.org/10.48550/arxiv.2011.08891) Gradients also measure local sensitivity rather than causal contribution, and the averaged gradient can miss localized evidence or weaken when the model is saturated.[17](https://arxiv.org/pdf/2608.12299.pdf) When the network output is near saturation, gradients can be practically zero and distort the heatmap; this workaround applies to implementations that differentiate post-softmax probabilities, since standard Grad-CAM already uses the pre-softmax score, and removing the softmax does not generally resolve gradient saturation upstream in the network.[21](https://ar5iv.labs.arxiv.org/html/2205.10900) HiResCAM, by contrast, has an L2 distance of always 0 between its explanation and the ground-truth locations used on PASCAL VOC 2012, while for Grad-CAM the distance is always nonzero.[3](https://doi.org/10.48550/arxiv.2011.08891)\n\n**Alternatives.** Occlusion-based perturbation, as used by Zeiler and Fergus, classifies images with patches occluded and is typically an order of magnitude more expensive than Grad-CAM's single forward and partial backward pass.[1](https://arxiv.org/abs/1610.02391) In a 2025 benchmark across SVHN, CIFAR-10, and Imagenette with VGG16BN, ResNet50, and DenseNet121, Grad-CAM and Grad-CAM++ were the fastest methods evaluated, though the choice of occlusion color in deletion metrics affected fidelity comparisons.[2](https://www.nature.com/articles/s41598-025-25839-y) [Evaluation](https://www.edgepedia.org/evaluation) across CAM methods remains fragmented, with faithfulness, localization, robustness, cost, and human trust often measured under different protocols; a method-centered review of 57 CAM papers covers gradient-based, gradient-free, high-resolution, weakly supervised, transformer, and foundation-model-era methods, reflecting the field's shift beyond CNNs.[17](https://arxiv.org/pdf/2608.12299.pdf)\n\n## References\n\n1. [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization (arXiv preprint, first posted October 2016)](https://arxiv.org/abs/1610.02391)\n2. [A comparative evaluation of explainability techniques for image data (Scientific Reports)](https://www.nature.com/articles/s41598-025-25839-y)\n3. [Draelos, Rachel Lea, Carin, Lawrence (2020). Use HiResCAM instead of Grad-CAM for faithful explanations of convolutional neural networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2011.08891)\n4. [jacobgil/pytorch-grad-cam](https://github.com/jacobgil/pytorch-grad-cam)\n5. [Grad-CAM publication page, Facebook AI Research](https://ai.meta.com/research/publications/grad-cam-visual-explanations-from-deep-networks-via-gradient-based-localization/)\n6. [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization (IJCV, Springer)](https://link.springer.com/article/10.1007/s11263-019-01228-7)\n7. [Learning Deep Features for Discriminative Localization (Zhou et al., CVPR 2016)](https://openaccess.thecvf.com/content_cvpr_2016/papers/Zhou_Learning_Deep_Features_CVPR_2016_paper.pdf)\n8. [kazuto1011/grad-cam-pytorch grad_cam.py](https://github.com/kazuto1011/grad-cam-pytorch/blob/master/grad_cam.py)\n9. [Selvaraju, Ramprasaath R. and colleagues (2016). Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1610.02391)\n10. [Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization (IEEE Xplore record)](https://ieeexplore.ieee.org/document/8237336)\n11. [Springenberg, Jost Tobias and colleagues (2014). Striving for Simplicity: The All Convolutional Net. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1412.6806)\n12. [Zhou, Bolei and colleagues (2015). Learning Deep Features for Discriminative Localization. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1512.04150)\n13. [Chattopadhyay, Aditya and colleagues (2017). Grad-CAM++: Improved Visual Explanations for Deep Convolutional Networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1710.11063)\n14. [Wang, Haofan and colleagues (2019). Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1910.01279)\n15. [Fu, Ruigang and colleagues (2020). Axiom-based Grad-CAM: Towards Accurate Visualization and Explanation of CNNs. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2008.02312)\n16. [Sattarzadeh, Sam and colleagues (2021). Integrated Grad-CAM: Sensitivity-Aware Visual Explanation of Deep Convolutional Networks via Integrated Gradient-Based Scoring. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2102.07805)\n17. [Class Activation Mapping in Explainable Computer Vision: A Method-Centered Review of CNN, Transformer, and Foundation-Model-Era Visual Explanations](https://arxiv.org/pdf/2608.12299.pdf)\n18. [LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions (CVPR 2025)](https://openaccess.thecvf.com/content/CVPR2025/papers/Mehri_LibraGrad_Balancing_Gradient_Flow_for_Universally_Better_Vision_Transformer_Attributions_CVPR_2025_paper.pdf)\n19. [Mehri, Faridoun, Baghshah, Mahdieh Soleymani, Pilehvar, Mohammad Taher (2024). LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2411.16760)\n20. [Attention Guided CAM: Visual Explanations of Vision Transformer Guided by Self-Attention](https://arxiv.org/abs/2402.04563)\n21. [RSI-Grad-CAM: Visual Explanations from Deep Networks via Riemann-Stieltjes Integrated Gradient-based Localization](https://ar5iv.labs.arxiv.org/html/2205.10900)\n\n---\n*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Artificial intelligence and data › Machine learning and neural computation › Neural networks and deep learning › Titles A to H*\n\n*Initially written Sep 29, 2026 · Edited: Oct 15, 2026*\n\n*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*\n\nLicense: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license\n", "same_as": [], "url": "https://www.edgepedia.org/grad-cam", "markdown_url": "https://www.edgepedia.org/grad-cam.md", "license": { "name": "Edgepedia Community License 1.0", "url": "https://www.edgepedia.org/edgepedia/license", "summary": "Free with credit, commercial use included. AI training is open to everyone. For other uses, organizations over USD 100M in revenue or 100M monthly users license separately.", "spdx": "LicenseRef-Edgepedia-Community-1.0" }, "credit": "\"Grad-CAM\", Edgepedia (EdgeChat), https://www.edgepedia.org/grad-cam. Edgepedia Community License 1.0.", "credit_md": "\"[Grad-CAM](https://www.edgepedia.org/grad-cam)\", Edgepedia (EdgeChat), [https://www.edgepedia.org/grad-cam](https://www.edgepedia.org/grad-cam). [Edgepedia Community License 1.0](https://www.edgepedia.org/edgepedia/license).", "credit_html": "\"Grad-CAM\", Edgepedia (EdgeChat), https://www.edgepedia.org/grad-cam. Edgepedia Community License 1.0.", "speakable": "Grad-CAM is a visualization technique producing heatmaps of the image regions a convolutional neural network relied on for a prediction, introduced in 2016." }