Reproduce “FoV-Net: Rotation-Invariant CAD B-rep Learning via Field-of-View Ray Casting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Matteo Ballegeer, Dries F. Benoit
FoV-Net enables rotation-invariant 3D CAD B-rep learning through local reference frame UV-grids and field-of-view ray casting descriptors, maintaining over 96% accuracy on arbitrarily rotated classification tasks where baseline methods drop below 15% and showing improved data efficiency in segmentation.
Key insights
- Rotation sensitivity is a critical, unaddressed limitation in B-rep learning, causing severe performance drops (e.g., from 97% to 9% accuracy in classification) without specific countermeasures.
- Combining local rotation-invariant geometric features (LRF UV-grids) with global rotation-invariant contextual features (FoV descriptors) effectively solves the problem of rotation sensitivity while maintaining high performance.
- Intrinsic rotation invariance improves model generalization and data efficiency, reducing the need for extensive training data (e.g., requiring 20x fewer samples than baselines to achieve similar segmentation accuracy).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.