Vina-Embroidery: A multi-angle image dataset and pixel-level annotations for anomaly localization in industrial embroidery inspection

Description

The Vina-Embroidery dataset is specifically designed for multi-view industrial anomaly detection and defect localization in textile manufacturing. The main dataset contains 1,080 high-resolution RGB images (492 normal and 588 abnormal samples) captured from three distinct viewpoints: 90 degrees (orthographic), 45 degrees (moderate off-axis), and 60 degrees (steeper off-axis) to capture surface textures and topological shadow cues. Additionally, an auxiliary localization subset is provided in YOLO format, containing 954 images, 954 label files, and 1,883 precise polygon annotations covering four primary industrial defect classes: surface gloss defect, excess thread, embroidery defect, and broken thread. This dataset serves as a standardized benchmark for evaluating texture-aware anomaly localization and modern object detection models.

Steps to reproduce

Download and extract the dataset archive while preserving the original hierarchical directory structure. If the dataset is provided as separate archives, extract the raw image data, split files, and YOLO annotation subset into the same root directory. Refer to raw/vai_theu/manifest.csv for the image-level metadata table, including image IDs, relative image paths, view angles, normal/abnormal labels, sample IDs, defect type IDs, and group IDs. For anomaly detection experiments, use the predefined partition files in the splits/vai_theu directory. Three viewpoint configurations are provided: straight_only, straight_45, and straight_60. Each configuration contains split.csv and summary.json files. For instance-level defect localization or object detection experiments, use the auxiliary subset in yolo_annotations/merged_yolo. The YOLO class indices and dataset paths are provided in lab_mendeley.yaml, and the class descriptions are listed in classes_en.txt.

Institutions

• Ho Chi Minh City University of Technology

  Ho Chi Minh, Ho Chi Minh City

Categories

Licence