SegPCBX Dataset

SegPCBX

Description The SegPCBX dataset encompasses 2D reconstructed X-ray images derived from the most optimized X-ray scan, offering extensive research opportunities in the fields of PCB reverse engineering and assurance such as component segmentation, design extraction, layer identification etc. This initial release features a sample PCB board with meticulously annotated multiple layer images, categorized into six common PCB component classes: vias, pads, traces, PGAs, solder balls, and copper planes. The annotations are formatted in COCO JSON format which can be easily adapted in any benchmark machine learning and deep learning algorithms for component segmentation purpose. Each layer image contains around 400 annotated components/instances; for example, the first layer of PCB 9 includes 140 vias, 132 solder balls, 62 pads, 71 traces, and 2 copper planes. Ongoing efforts are underway to expand this database by collecting and annotating additional X-ray images.
Downloads SegPCBX