2021/06/02 by Pierre Gutierrez, Antoine Cordier, Gutierrez, Pierre +5
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #Anomaly Detection Techniques and Applications #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #I.2.10 #I.4.6 #I.4.8 #I.4.9 #I.5 #Image Processing Techniques and Applications #Industrial Vision Systems and Defect Detection #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.2106.01277
openalex publication_date 2021/06/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The use of deep features coming from pre-trained neural networks for\nunsupervised anomaly detection purposes has recently gathered momentum in the\ncomputer vision field. In particular, industrial inspection applications can\ntake advantage of such features, as demonstrated by the multiple successes of\nrelated methods on the MVTec Anomaly Detection (MVTec AD) dataset. These\nmethods make use of neural networks pre-trained on auxiliary classification\ntasks such as ImageNet. However, to our knowledge, no comparative study of\nrobustness to the low data regimes between these approaches has been conducted\nyet. For quality inspection applications, the handling of limited sample sizes\nmay be crucial as large quantities of images are not available for small\nseries. In this work, we aim to compare three approaches based on deep\npre-trained features when varying the quantity of available data in MVTec AD:\nKNN, Mahalanobis, and PaDiM. We show that although these methods are mostly\nrobust to small sample sizes, they still can benefit greatly from using data\naugmentation in the original image space, which allows to deal with very small\nproduction runs.\n