Zhenglin Chen

AD [8] 92.4 92.4 89.6 98.1 38.0 42.6 91.8 DeSTSeg [45] 88.9 89.0 85.2 96.1 39.6 43.4 67.4 UniAD [41] 88.8 90.8 85.8 98.3 33.7 39.0 85.5 ReContrast [16] 95.5 96.4 92.0 98.5 47.9 50.6 91.9 DiAD [19] 86.8 88.3 85.1 96.0 26.1 33.0 75.2 ViTAD [44] 90.5 91.7 86.3 98.2 36.6 41.1 85.1 MambaAD [18] 94.3 94.5 89.4 98.5 39.4 44.0 91.0 Dinomaly [17] 98.7 98.9 96.2 98.7 53.2 55.7 94.5 StructCore (Ours, Base, 1%) 97.6 98.1 94.3 98.6 50.1 53.4 95.6 StructCore (Ours, Hybrid, 1%) 98.4 98.5 95.7 98.6 50.1 53.4 95.6 Table 4: Scalability with memory-bank size on MVTec AD (RTX 3090, batch=16). “Extract” includes DINOv2 feature extraction and random projection; “kNN” is FAISS search; “Post” includes Gaussian blur and StructCore scoring. Coreset ratio controls the memory-bank size. Coreset Extract (ms) kNN (ms) Post (ms) Total (ms) FPS GPU peak (MB) 1% 9.60 0.20 0.32 10.12 98.85 2333 5% 9.68 0.68 0.31 10.67 93.71 2459 10% 9.73 1.29 0.27 11.30 88.53 2619 4.3 Ablation Study Sensitivity to the hybrid weight λ\lambda. StructCore uses the hybrid weight λ\lambda to control the contribution of the structural calibration term. We perform a sensitivity analysis over λ∈{0.05,0.10,0.15,0.20}\lambda\in\{0.05,0.10,0.15,0.20\} across all categories. While the optimal λ\lambda may vary by category, the overall mean image-level AUROC remains essentially unchanged (around 99.6%) across these values, indicating that StructCore is not overly sensitive to λ\lambda in this range. Accordingly, we use a single fixed default λ=0.05\lambda{=}0.05 in all experiments without category-wise tuning. structural descriptor ϕ​(S)\phi(S). StructCore hinges on a compact structural descriptor ϕ​(S)\phi(S) that summarizes an anomaly score map beyond its maximum response. Because ϕ​(S)\phi(S) is intentionally low-dimensional and training-free, it is important to verify that (i) each component captures complementary information that max pooling discards, and (ii) the full