Papers › Informative knowledge distillation for image anomaly segmentation

Informative knowledge distillation for image anomaly segmentation

19 Jul 2022Knowledge-Based Systems 2022 7archive 2025-07-28

Yunkang Cao, Qian Wan, Weiming Shen, Liang Gao

Unsupervised anomaly segmentation methods based on knowledge distillation have recently been developed and show superior segmentation performance. However, rare attention has been paid to the overfitting problem caused by the inconsistency between the capacity of the neural network and the amount of knowledge in this scheme. This paper proposes a novel method named Informative Knowledge Distillation (IKD) to address the overfitting problem by increasing knowledge and offering a strong supervisory signal. Technically, a novel Context Similarity Loss (CSL) is proposed to capture context information from normal data manifolds. Besides, a novel Adaptive Hard Sample Mining (AHSM) is proposed to encourage more attention on hard samples with valuable information. With IKD, informative knowledge can be distilled, so that the overfitting problem can be well mitigated, and the performance can be further increased. The proposed method achieves better results on several categories of the well-known MVTec AD dataset than state-of-the-art methods in terms of AU-ROC, achieving 97.81% overall in 15 categories. Extensive experiments on ablation studies are also conducted to show the effectiveness of IKD in alleviating the overfitting problem.

PaperPDFCode

Code

caoyunkang/IKD officialpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Anomaly DetectionAnomaly SegmentationKnowledge Distillation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Anomaly Detection MVTec AD IKD Segmentation AUPRO 92.55 #128 of 148 Archive leaderboard report
Anomaly Detection MVTec AD IKD Segmentation AUROC 97.81 #128 of 148 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Knowledge Distillation

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections