PKU-GoodsAD: A Supermarket Goods Dataset for Unsupervised Anomaly Detection and Segmentation

TL;DR

This work introduces GoodsAD, a large-scale, high-resolution supermarket goods dataset designed for unsupervised visual anomaly detection and segmentation. It enables concurrent evaluation of image-level classification and pixel-level localization under a setting where only defect-free images are used for training. The authors benchmark a broad spectrum of SOTA approaches across pre-trained features, pseudo-anomalies, and generative models, revealing that existing methods underperform in real-world goods scenarios and that object misalignment and appearance variation pose substantial challenges. PatchCore emerges as the strongest performer on GoodsAD, yet overall accuracy remains insufficient for practical deployment, underscoring the need for further advances in unsupervised anomaly detection for unmanned store environments.

Abstract

Visual anomaly detection is essential and commonly used for many tasks in the field of computer vision. Recent anomaly detection datasets mainly focus on industrial automated inspection, medical image analysis and video surveillance. In order to broaden the application and research of anomaly detection in unmanned supermarkets and smart manufacturing, we introduce the supermarket goods anomaly detection (GoodsAD) dataset. It contains 6124 high-resolution images of 484 different appearance goods divided into 6 categories. Each category contains several common different types of anomalies such as deformation, surface damage and opened. Anomalies contain both texture changes and structural changes. It follows the unsupervised setting and only normal (defect-free) images are used for training. Pixel-precise ground truth regions are provided for all anomalies. Moreover, we also conduct a thorough evaluation of current state-of-the-art unsupervised anomaly detection methods. This initial benchmark indicates that some methods which perform well on the industrial anomaly detection dataset (e.g., MVTec AD), show poor performance on our dataset. This is a comprehensive, multi-object dataset for supermarket goods anomaly detection that focuses on real-world applications.

Figures (6)

• Figure 1: Virtual scene of unmanned supermarket shopping based on computer vision technology.
• Figure 2: Example images for six categories of the GoodsAD dataset. The first to sixth rows show bottled drink, canned drink, bottled food, boxed food, packaged food and boxed cigarette, respectively. Each category contains multiple items. The green pane on the left shows anomaly-free images, and the red pane on the right shows corresponding anomalous examples. Anomalous regions are highlighted in red polygons.
• Figure 3: Statistics on the ratio of the number of pixels in the anomalous region to the entire image. Different coloured circles represent different types of anomalous samples.
• Figure 4: Inference speed and the model parameters stored for each method.
• Figure 5: Visual examples of anomaly localization by five different methods on the GoodsAD dataset. GT denotes ground truth. From left to right, each of the two columns is the images of categories drink_bottle, drink_can, food_bottle, food_box, food_package and cigarette_box.