Mixed supervision for surface-defect detection: from weakly to fully supervised learning (2104.06064v3)

Abstract: Deep-learning methods have recently started being employed for addressing surface-defect detection problems in industrial quality control. However, with a large amount of data needed for learning, often requiring high-precision labels, many industrial problems cannot be easily solved, or the cost of the solutions would significantly increase due to the annotation requirements. In this work, we relax heavy requirements of fully supervised learning methods and reduce the need for highly detailed annotations. By proposing a deep-learning architecture, we explore the use of annotations of different details ranging from weak (image-level) labels through mixed supervision to full (pixel-level) annotations on the task of surface-defect detection. The proposed end-to-end architecture is composed of two sub-networks yielding defect segmentation and classification results. The proposed method is evaluated on several datasets for industrial quality inspection: KolektorSDD, DAGM and Severstal Steel Defect. We also present a new dataset termed KolektorSDD2 with over 3000 images containing several types of defects, obtained while addressing a real-world industrial problem. We demonstrate state-of-the-art results on all four datasets. The proposed method outperforms all related approaches in fully supervised settings and also outperforms weakly-supervised methods when only image-level labels are available. We also show that mixed supervision with only a handful of fully annotated samples added to weakly labelled training images can result in performance comparable to the fully supervised model's performance but at a significantly lower annotation cost.

Mixed Supervision for Surface-Defect Detection: From Weakly to Fully Supervised Learning

The rise of Industry 4.0 emphasizes the importance of automated visual inspection in manufacturing, where deep learning has progressively replaced traditional machine vision methods. This paper addresses the pervasive challenge in deep-learning-based anomaly detection: the scarcity and high cost of obtaining precise, fully-annotated data. It proposes a novel approach leveraging mixed supervision to optimize surface-defect detection, a critical component in industrial quality assurance.

The paper introduces a deep-learning architecture that harmonizes weakly supervised with fully supervised learning by employing annotations with varying levels of detail. The model integrates two interconnected sub-networks designed for defect segmentation and classification. This unification enables effective exploitation of diverse datasets: those only with image-level labels and others with pixel-level annotations.

The proposed methodology is rigorously evaluated across multiple datasets, such as KolektorSDD, DAGM, Severstal Steel Defect, and a newly introduced KolektorSDD2 dataset, which comprises over 3000 images from a real-world industrial context. Notably, the model delivers state-of-the-art results, outperforming both fully supervised and weakly supervised methods when mixed supervision is applied judiciously.

Mixed supervision demonstrates its efficacy by drastically reducing annotation costs while maintaining performance at par with fully supervised models. This is achieved by integrating a minimal number of fully annotated samples into a predominantly weakly annotated dataset. Such an approach reshapes the cost-performance landscape, making complex annotation tasks more manageable and economically feasible for industrial applications.

This paper also highlights the implications of utilizing varied annotation granularities. By cautiously balancing the annotation burden against the quality of the defect detection, the research not only offers a practical solution to the prevalent issue of expensive data labeling in industrial settings but also contributes theoretically to the understanding of supervision levels in machine learning.

In conclusion, as deep learning continues to permeate across industries, the implementation of mixed supervision conveys substantial promise. It opens pathways for further exploration into balancing data quality with computational efficiency, potentially influencing future developments and methodologies in anomaly detection and beyond. This work suggests that forthcoming AI models could increasingly integrate varying degrees of data supervision to optimize performance while minimizing the high costs associated with exhaustive data labeling.