Architecture-Aware Explanation Auditing for Industrial Visual Inspection

2026-05-13

AI Summary

This work addresses the discrepancy between visually plausible heatmaps and the actual decision mechanisms of deep models in industrial visual inspection. The authors propose an architecture-aware explanation auditing protocol grounded in the “native readout hypothesis,” which quantitatively evaluates explanation fidelity using Deletion AUC under zero-fill and blur-fill perturbations. Experiments across diverse architectures—including ViT, Swin, ResNet18+CBAM, and DenseNet121—and explanation methods such as Attention Rollout, Grad-CAM, and RISE on the WM-811K dataset reveal that explanation fidelity hinges on a model’s native readout structure rather than its architectural family. Notably, Swin-Tiny’s spatially hierarchical feature maps render it compatible with Grad-CAM, while RISE, as a model-agnostic method, achieves the highest fidelity overall. These findings underscore that fidelity is a joint property of the (model, explainer, perturbation operator) triplet, necessitating co-design of explanation pathways with model architecture and validation via quantitative metrics.

Abstract

Industrial visual inspection systems increasingly rely on deep classifiers whose heatmap explanations may appear visually plausible while failing to identify the image regions that actually drive model decisions. This paper operationalizes an architecture-aware explanation audit protocol grounded in the native-readout hypothesis: the perturbation-based faithfulness of an explanation method is bounded by its structural distance from the model's native decision mechanism. On WM-811K wafer maps (9 classes, 172k images) under a three-seed zero-fill perturbation protocol, ViT-Tiny + Attention Rollout attains Deletion AUC 0.211 against 0.432-0.525 for Swin-Tiny / ResNet18+CBAM / DenseNet121 + Grad-CAM (abs(Cohen's d) > 1.1), despite lower classification accuracy. Swin-Tiny disentangles architecture family from readout structure: despite being a Transformer, its spatial feature-map hierarchy makes it Grad-CAM compatible, showing that the operative factor is readout structure rather than architecture family. A model-agnostic control (RISE) compresses all families to Deletion AUC about 0.1, indicating the gap arises from the explainer pathway; notably, RISE outperforms all native methods, so native readout is a compatibility principle rather than an optimality guarantee. A blur-fill sensitivity analysis shows that the family ordering reverses under a different perturbation baseline, reinforcing that faithfulness rankings are joint properties of (model, explainer, perturbation operator) triples. An exploratory boundary-condition study on MVTec AD (pretrained models) indicates that audit results are dataset/task dependent and identifies conditions requiring qualification. The protocol yields actionable guidance: explanation pathways should be co-designed with model architectures based on readout structure, and deployed heatmaps should be accompanied by quantitative faithfulness metrics.

Problem

• explanation faithfulness
• visual inspection
• model interpretability
• heatmap evaluation
• architecture-aware auditing

Innovation

• architecture-aware explanation
• native-readout hypothesis
• faithfulness auditing
• readout structure
• perturbation-based evaluation