Class Activation Mapping in Explainable Computer Vision: A Method-Centered Review of CNN, Transformer, and Foundation-Model-Era Visual Explanations

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: AmirHossein Eshghi, Hamid Saadatfar, Seyyed Ali Hoseini, AmirMohsen Eshghi, Siavash Arjomand Bigdel
• Venue: arXiv: cs.CV
• Published: 2026-08-12
• Collected: 2026-08-13
• URL: http://arxiv.org/abs/2608.12299v1
• PDF: https://arxiv.org/pdf/2608.12299v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

This research paper provides a comprehensive review of various visual explanation methods used in explainable artificial intelligence, specifically focusing on Class Activation Mapping (CAM). The paper discusses the evolution of CAM from its initial formulation in 2016, covering gradient-based post-hoc explanations, high-resolution upscaling, transformer token attribution, and foundation-model-era approaches. The review also presents a taxonomy to categorize methods based on their attribution mechanism, architectural dependence, and evaluation objective.

What the Document Says

• The paper discusses Class Activation Mapping (CAM), a widely used visual explanation method in explainable AI.
• CAM converts internal model evidence into a heatmap that highlights relevant image regions, convolutional channels, tokens, or patches for a target class or concept.
• Since its first formulation in 2016, CAM-style methods have expanded beyond global-average-pooled CNN classifiers to include various other techniques.
• The review covers a strict corpus of 57 method-centered papers published from 2016 onwards.
• The paper develops a taxonomy separating methods by attribution mechanism, architectural dependence, and evaluation objective.
• The review discusses gradient-based CAMs, recent and hybrid CAM-style methods, and model-based or architecture-aware methods.

Key Concepts Mentioned

• Class Activation Mapping (CAM)
• Explainable Artificial Intelligence
• Visual Explanations
• CNN
• Transformer
• Foundation Models

Claims Made by the Document

• The field of visual explanations has evolved significantly since the first CAM formulation in 2016.
• Across the reviewed corpus, the main trend is shifting from explaining one class score in one low-resolution CNN layer towards comparative, multi-layer, probabilistic, token-aware, and foundation-model-aware explanations.

Evidence Used

• The text discusses various techniques that have emerged since the first CAM formulation.
• The paper emphasizes the shift in explanations towards more complex methods.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Class activation mapping (CAM) is one of the most widely used visual explanation families in explainable artificial intelligence. Its purpose is intuitive: it converts internal model evidence into a heatmap that highlights the image regions, convolutional channels, tokens, or patches that support a target class or concept. Since the first CAM formulation in 2016, the field has moved far beyond global-average-pooled CNN classifiers. CAM-style methods now include gradient-based post-hoc explanations, gradient-free score and ablation methods, high-resolution upscaling, weakly supervised localization and segmentation, transformer token attribution, causal and debiasing methods, and foundation-model-era approaches that use CLIP, DINO, SAM, or feature-distribution comparisons. This review synthesizes a strict corpus of 57 method-centered papers published from 2016 onward. The paper develops a taxonomy that separates methods by attribution mechanism, architectural dependence, and evaluation objective. It then reviews gradient-based CAMs, recent and hybrid CAM-style methods, and model-based or architecture-aware methods. Across the corpus, the main trend is clear: the field is shifting from explaining one class score in one low-resolution CNN layer toward comparative, multi-layer, probabilistic, token-aware, and foundation-model-aware explanations. At the same time, evaluation remains fragmented. Faithfulness, localization, robustness, computational cost, and human trust are often measured with different protocols. The review therefore emphasizes not only what each method contributes, but also which gap it leaves open and which later methods attempt to close that gap.