What is it about?
This work presents a new deep learning model called BRepFormer that uses transformers to recognize geometric features in CAD models represented by boundary representation (B-rep). Unlike previous methods that lose important geometric or topological details, BRepFormer directly processes B-rep data, encoding both geometric and topological features. It uses a novel attention mechanism to fuse these features and improves the accuracy of identifying both machining features and complex CAD structures. Additionally, the paper introduces a new large dataset of complex B-rep models to better evaluate the approach.
Featured Image
Photo by Vertex Designs on Unsplash
Why is it important?
Recognizing geometric features in CAD models is essential for bridging design and manufacturing automation, but current methods struggle with complex shapes and topologies, often losing critical details. BRepFormer overcomes these challenges by effectively integrating geometric and topological information within a transformer framework, achieving state-of-the-art accuracy on multiple benchmark datasets. This advancement can significantly enhance automated manufacturing processes, reduce human error, and improve the efficiency of CAD model analysis.
Perspectives
From my perspective, this research pushes forward the application of transformers in 3D CAD analysis, especially in handling the complex topology of B-rep models. The introduction of the CBF dataset further provides a valuable resource for future studies in geometric feature recognition. The integration of global and local features via the attention bias mechanism seems particularly promising for other 3D model understanding tasks as well.
Yong Kang Dai
Read the Original
This page is a summary of: BRepFormer: Transformer-Based B-rep Geometric Feature Recognition, June 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3731715.3733283.
You can read the full text:
Contributors
The following have contributed to this page
