PCB-IND: Industrial Printed Circuit Board Surface Defect Dataset for Object Detection

PCB-IND is an industrial printed circuit board (PCB) surface defect dataset for object detection. The dataset contains 4,789 ROI images collected from a real automated optical inspection (AOI) production line and covers eight defect categories: mouse_bite, missing_copper, scratch, spurious_copper, copper_burr, stain, short, and open. The dataset is split into training (3,833 images), validation (478 images), and test (478 images) subsets. All image patches are provided at a fixed resolution of 300 × 300 pixels. Verified annotations are released in YOLO, PASCAL VOC, and MS COCO formats to support different object detection frameworks. To improve release consistency and reproducibility, the updated version includes a machine-readable label map file (classes.json) and a revised README file that explicitly documents the official class definitions, dataset organization, and annotation formats. The dataset is primarily intended for AOI candidate verification and patch-level defect detection under real industrial imaging conditions. License: CC BY 4.0.