Geometry-Aligned Semantic Matching for Cross-Modal Planar Image Registration
Paper • 2610.03167 • Published
Project Page · Online Demo · Paper · Code
PCB-Layout contains 668 pairs of PCB photographs and circuit layout images, with each pair manually aligned to establish spatial correspondence.
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Rows: PCB · Layout · Animated overlay (PCB ↔ blend ↔ Layout)
![]() |
Capture: HIKROBOT MV-CU060-10UC · 3072 × 2048 · 10–50 mm zoom lens · polarized illumination. Collection: 100 PCB surfaces, seven images per surface across dark, normal, and bright illumination; 668 pairs retained after filtering. Alignment: Each pair was annotated with a homography and verified through manual inspection and reprojection error. |
PCB-Layout/
├── PCB/ # 668 PCB photographs
│ └── <sample_id>.png
└── Layout/ # 668 corresponding layouts
└── <sample_id>.png
Paired images share the same filename and resolution. All images are RGB PNGs.
If you find this dataset helpful, please cite our paper:
@article{wang2026cdpm,
title={Geometry-Aligned Semantic Matching for Cross-Modal Planar Registration},
author={Wang, Zhiwei and He, Defeng and Li, Yuxing and Zhu, Meilu and Lam, Edmund Y.},
journal={arXiv},
year={2026}
}