Architecture-Aware Explanation Auditing for Industrial Visual Inspection
For practitioners deploying deep classifiers in industrial visual inspection, this work provides actionable guidance to co-design explanation pathways with model architectures based on readout structure and to accompany heatmaps with quantitative faithfulness metrics.
The paper proposes an architecture-aware explanation audit protocol for industrial visual inspection, showing that perturbation-based faithfulness of explanation methods is bounded by their structural distance from the model's native decision mechanism. On WM-811K wafer maps, ViT-Tiny + Attention Rollout achieves Deletion AUC 0.211 versus 0.432-0.525 for other methods, despite lower accuracy, and RISE outperforms all native methods, indicating native readout is a compatibility principle rather than optimality guarantee.
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.