Reproduce “Circuit Diagram Retrieval Based on Hierarchical Circuit Graph Representation” on OpenResearch

About this paper

Ming Gao, Ruichen Qiu, Zeng Hui Chang, Kanjian Zhang, Haikun Wei, Hong Cai Chen

Researchers from Southeast University developed a system for retrieving circuit diagrams by transforming them into hierarchical graph representations and comparing them using graph edit distance. This method achieved an AP_retrieval@5 of 0.881 with an average retrieval time of 57.86 seconds, significantly outperforming traditional image-based retrieval techniques.

Key insights

• Converting circuit diagrams into graph representations effectively captures their inherent topological and functional information, which is critical for accurate retrieval and cannot be reliably obtained from purely visual similarity.
• Implementing a hierarchical retrieval strategy addresses the computational expense of Graph Edit Distance (GED), enabling high retrieval accuracy while maintaining practical processing times for large databases.
• Combining state-of-the-art deep learning (GAM-YOLO) for object recognition with graph-based techniques provides a robust and accurate pipeline for transforming visual circuit diagrams into a semantically rich, retrievable format.

Summary by alphaXiv.