Rectifying geometry–topology inconsistency in ComplexGen CAD reconstructions

Determine an effective method to rectify the inconsistency between predicted geometry and topology in CAD models generated by ComplexGen, so that the resulting boundary representations are valid and consistent. Specifically, devise a procedure that ensures the surfaces, edges, and corners produced by ComplexGen adhere to geometric-topological compatibility constraints and yield valid CAD models.

Background

The paper compares the proposed Point2CAD pipeline with ComplexGen for reconstructing CAD models from point clouds. While ComplexGen can generate surfaces, edges, and corners along with topological relationships, the authors observe that its outputs are often invalid because the predicted geometry and topology are inconsistent.

This inconsistency presents a practical challenge in CAD reconstruction pipelines: even when individual elements are predicted, ensuring their mutual compatibility is necessary to produce usable boundary-representation (B-rep) models. The authors note they are not aware of an easy way to fix this issue and suggest a possible workaround—refitting and intersecting ComplexGen’s surfaces with their own pipeline—underscoring the need for a direct and general rectification method.

References

However, it should be noted that the models generated by ComplexGen are often invalid, in the sense that the predicted geometry and topology are inconsistent. We are not aware of an easy way to rectify this issue. A possible solution could be only to utilize the surfaces generated by ComplexGen and apply our method to refit and intersect them.

Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds  (2312.04962 - Liu et al., 2023) in Section 4 (Experiments), Subsection "Reconstructed CAD evaluation"