PCB-SAID: A Low-Cost Camera-Based Dataset for Few-Shot SMD Assembly Inspection

Raffaele Mineo, Amelia Sorrenti, Robin Faro, Gabriele Mineo, Francesco Cancelliere, Alberto Faro; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 1362-1368

Automated Optical Inspection (AOI) of surface-mounted device (SMD) assemblies is critical for ensuring quality and yield in electronics manufacturing, yet current systems rely on expensive hardware and handcrafted algorithms that struggle with novel layouts and rare defect modes. We present PCB-SAID (Printed Circuit Board - SMD Assembly Inspection Dataset), the first open, few-shot SMD inspection dataset comprising 175 RGB images annotated with 66 fine-grained classes, spanning well-soldered, mis-soldered (misaligned, lifted, rotated), missing, and short-circuit conditions, using a rigorous two-stage expert pipeline (bounding-box localization followed by polygonal mask refinement). We evaluate five state-of-the-art architectures (Fast R-CNN, DETR, YOLOv11, YOLOv12, SAM) under both full-data and few-shot (1, 5 shots) regimes, reporting mAP@[.50:.95], mAP@.50, precision, and recall. Our results reveal optimal performance gaps on rare and subtle anomalies, underscoring challenges in class imbalance and domain shift. The dataset will be made publicly available upon request to support further research in this area.