Paper Overview
- Field: Computer Vision
- Authors: AmirHossein Eshghi, Hamid Saadatfar, Seyyed Ali Hoseini
- Published: 2026-08-13
- arXiv: 2508.03414
- Gradient-based post-hoc explanations
- Gradient-free scores and ablation methods
- High-resolution upsampling techniques
- Weakly supervised localization and segmentation
- Transformer token attribution
- Causal and debiasing approaches
- Foundation-model-era methods using CLIP, DINO, SAM, or feature-distribution comparison
Key Points
Class activation mapping (CAM) is one of the most widely used visual explanation families in explainable AI. Its goal is intuitive: convert internal model evidence into heatmaps that highlight image regions, convolution channels, tokens, or patches supporting a target class or concept.
Since the first CAM formulation in 2016, the field has expanded far beyond global-average-pooling CNN classifiers. CAM-style methods now include:
Survey Structure
The survey synthesizes 57 method-focused studies published since 2016. The authors:
1. Develop a taxonomy that separates methods by attribution mechanism, architectural dependency, and evaluation target 2. Review gradient-based CAM, recent and hybrid CAM-style methods, and model- or architecture-aware approaches
Trends and Open Problems
A dominant trend is clear: the field is shifting from explaining a single class score in a single-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 typically measured under different protocols. The survey therefore emphasizes not only each method's contributions, but also the gaps it leaves and how subsequent methods attempt to fill them.
--- *Auto-collected on 2026-08-14*