How to Evaluate and Refine your CAM

2026-05-14

AI Summary

Existing Class Activation Mapping (CAM) methods suffer from unreliable evaluation due to the absence of ground-truth attribution labels and produce low-resolution attribution maps that limit interpretability. To address these issues, this work constructs the first synthetic dataset equipped with ground-truth attribution labels, enabling systematic benchmarking of current CAM evaluation metrics. The study further proposes RefineCAM, a high-resolution attribution method that enhances explanatory detail through multi-layer feature aggregation. Additionally, a composite evaluation metric, ARCC, is introduced to more accurately assess attribution fidelity. Experimental results demonstrate that RefineCAM significantly outperforms existing approaches on the ARCC metric, confirming its superior performance in both attribution fidelity and fine-grained visual explanation.

Abstract

Class attribution maps (CAMs) provide local explanations for the decisions of convolutional neural networks. While widely used in practice, the evaluation of CAMs remains challenging due to the lack of ground-truth explanations, making it difficult to evaluate the soundness of existing metrics. Independently, most commonly used CAM methods produce low-resolution attribution maps, which limits their usefulness for detailed interpretability. To address the evaluation challenge, we introduce a synthetic dataset with ground-truth attributions that enables a rigorous comparison of CAM evaluation metrics. Using this dataset, we analyze existing metrics and propose ARCC, a new composite metric that more reliably identifies faithful explanations. To address the low resolution issue, we introduce RefineCAM, a method that produces high-resolution attribution maps by aggregating CAMs across multiple network layers. Our results show that RefineCAM consistently outperforms existing methods according to the proposed evaluation.

Problem

• Class Activation Maps
• Explanation Evaluation
• Ground-truth Attribution
• Low-resolution Attribution
• Interpretability

Innovation

• Class Activation Maps
• ARCC
• RefineCAM
• ground-truth attributions
• high-resolution attribution