3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly

TL;DR

This work addresses the need for realistic industrial anomaly detection benchmarks by introducing 3CAD, a large-scale real-world 3C dataset with pixel-level defect annotations across eight part types and the possibility of multiple defects per image. To tackle the challenges of real-world defects, the authors propose CFRG, a coarse-to-fine localization framework with Recovery Guidance that combines heterogeneous knowledge distillation, a recovery network, and a segmentation module to achieve precise pixel-level localization. Empirical results show CFRG delivers competitive performance on 3CAD and reveals substantial gaps between real-world data and existing benchmarks, highlighting the task's difficulty and the potential for continued advances in unsupervised anomaly detection for 3C products. The dataset and framework together provide a practical benchmark and methodological direction to drive robust anomaly detection in real manufacturing environments.

Abstract

Industrial anomaly detection achieves progress thanks to datasets such as MVTec-AD and VisA. However, they suffer from limitations in terms of the number of defect samples, types of defects, and availability of real-world scenes. These constraints inhibit researchers from further exploring the performance of industrial detection with higher accuracy. To this end, we propose a new large-scale anomaly detection dataset called 3CAD, which is derived from real 3C production lines. Specifically, the proposed 3CAD includes eight different types of manufactured parts, totaling 27,039 high-resolution images labeled with pixel-level anomalies. The key features of 3CAD are that it covers anomalous regions of different sizes, multiple anomaly types, and the possibility of multiple anomalous regions and multiple anomaly types per anomaly image. This is the largest and first anomaly detection dataset dedicated to 3C product quality control for community exploration and development. Meanwhile, we introduce a simple yet effective framework for unsupervised anomaly detection: a Coarse-to-Fine detection paradigm with Recovery Guidance (CFRG). To detect small defect anomalies, the proposed CFRG utilizes a coarse-to-fine detection paradigm. Specifically, we utilize a heterogeneous distillation model for coarse localization and then fine localization through a segmentation model. In addition, to better capture normal patterns, we introduce recovery features as guidance. Finally, we report the results of our CFRG framework and popular anomaly detection methods on the 3CAD dataset, demonstrating strong competitiveness and providing a highly challenging benchmark to promote the development of the anomaly detection field. Data and code are available: https://github.com/EnquanYang2022/3CAD.

Figures (9)

• Figure 1: Comparison of previous anomaly detection distillation paradigm with our paradigm. First row: Left: reverse distillation; Right: our proposed paradigm.
• Figure 2: 3CAD dataset samples. The first row shows normal images, while the second row displays defective images.
• Figure 3: Statistics of the proposed 3CAD dataset: a) Defect area ratio. b) Aspect ratio of the minimum bounding rectangle for the defect area.
• Figure 4: The proposed CFRG framework comprises two components: 1) a distilled localization network and 2) a refined segmentation network with restored hints. During training, in the first stage, $x_a$ and $x_n$ are input into the teacher network, while $x_a$ is input into the student network, and the distillation loss between the teacher and the student is calculated. In the second stage, the teacher's features are weighted using the first-stage localization weights and the recovery branch's hint weights, then input into the segmentation network. During testing, the recovery branch generates the localization result from the input and $\{F_{i}^{r}\}_{i=1}^{K}$, which is then added to the output $S_{out}$ of the segmentation network to obtain the final anomaly map.
• Figure 5: Qualitative illustration on 3CAD dataset.