← Back to Computer Vision
cs.CV

How a reference image reveals PCB defects hidden in plain sight

Vinay Edula, Nilesh Badwe, Priyanka Bagade

May 30, 2026

PCB defect detection fails on tiny flaws buried in complex circuit patterns. RefDiffNet compares a defective board against a defect-free reference image to highlight structural changes, then feeds the enhanced result to any existing detector. On HRIPCB and DeepPCB benchmarks, it delivers 18% mAP improvement across YOLOv8–v26, RT-DETR, and Faster R-CNN while adding only 0.004M parameters—less than 0.25% overhead.
Published as RefDiffNet: Learning to Expose Subtle PCB Defects Before Detection arXiv:2606.00852
Read the original paper →