Real world CAD hole detection methods compared for accuracy and reliability

A Real-World CAD Segmentation Benchmark: Comparing Rule-Based and Learning-Based Approaches through Hole Detection

Computer Vision and Pattern Recognition

Summary

Detecting holes in computer-aided design (CAD) models is important for engineering tasks but existing datasets are mostly artificial or automatically labeled, which may not work well in real situations. The authors created a new dataset with human-verified hole annotations on real CAD models and developed two rule-based methods and trained learning-based methods to detect holes. They found that rule-based methods are reliable without needing training data but differ in detecting more holes or avoiding false alarms, while learning methods work well with enough training data but are less reliable when the data changes. This study helps understand the trade-offs between approaches for better real-world CAD segmentation.

What this means in practice

  • For mechanical engineers: Improve hole detection in CAD for assembling and manufacturing processes using methods validated on real engineering models.
  • For cad software developers: Incorporate robust hole detection algorithms into CAD tools that maintain performance without extensive retraining on new datasets.$Commercial implications: Enables selling enhanced CAD software with reliable feature recognition for engineering customers needing precision modeling.

Authors

Kai Schütte, Edgar Heinert, Matthias Rottmann

Abstract

CAD segmentation datasets are predominantly synthetic or automatically labelled, leaving it unclear how well existing methods transfer to real-world engineering tasks. Fastener-relevant hole detection represents one such task, requiring human annotation and exhibiting greater geometric complexity than commonly used benchmarks. We address this gap through three main contributions. First, we develop two task-specific rule-based methods, one graph-wire-search and one multi-view based, and complement them with a human-annotated benchmark dataset derived from engineering CAD data that also enables training of UV-Net and AAGNet, representative learning-based approaches for B-Rep segmentation. Second, we conduct an extensive experimental study including parameter optimisation, evaluation on our benchmark dataset, cross-domain generalisation on a synthetic dataset, and an assessment of the feasibility of the approaches at varying dataset sizes and IoU thresholds. Third, we provide evidence-based insights into the strengths and limitations of different approaches: rule-based methods generalise well without training data, with the graph-wire-search method favouring recall and the multi-view method favouring precision, whereas learning-based methods achieve strong performance with sufficient annotations but exhibit reduced robustness under domain shift. These findings highlight trade-offs between data requirements, and robustness, and provide guidance not only for hole detection, but more broadly for real-world CAD segmentation tasks.
View PDFOpen arXiv