Symmetrization of 3D Generative Models

TL;DR

This work introduces a data-centric approach to enforce reflectional symmetry in 3D generative models by training on half-objects and reconstructing full shapes via reflection. By evaluating symmetry through a Chamfer Distance-based protocol and distributional metrics like FPD on ShapeNet Airplane, Car, and Chair, the authors show that half-object training improves geometric symmetry without altering model architectures. The method demonstrates that symmetry priors can be learned from data alone, though some trade-offs in local detail fidelity remain, and it lays groundwork for extending symmetry-aware training to broader models. Overall, the study highlights a practical, architecture-agnostic strategy for instilling structural priors in 3D shape generation with potential broad impact on downstream 3D vision tasks.

Abstract

We propose a novel data-centric approach to promote symmetry in 3D generative models by modifying the training data rather than the model architecture. Our method begins with an analysis of reflectional symmetry in both real-world 3D shapes and samples generated by state-of-the-art models. We hypothesize that training a generative model exclusively on half-objects, obtained by reflecting one half of the shapes along the x=0 plane, enables the model to learn a rich distribution of partial geometries which, when reflected during generation, yield complete shapes that are both visually plausible and geometrically symmetric. To test this, we construct a new dataset of half-objects from three ShapeNet classes (Airplane, Car, and Chair) and train two generative models. Experiments demonstrate that the generated shapes are symmetrical and consistent, compared with the generated objects from the original model and the original dataset objects.

Figures (5)

• Figure 1: (a) Reference shapes from the ShapeNet dataset: Airplane, Car, and Chair. (b) Unconditional shape generation with 2,048 points for each class-specific model. Red circles highlight missing, incomplete and deformed regions.
• Figure 2: Reflection symmetry computation in ShapeNet - 3 classes.
• Figure 3: Symmetry computation using the proposed measurement protocol on 3D generative models (PVD, LION, XCube, and SLIDE 3D) across the three ShapenNet classes. Ground-truth results from ShapeNet are provided for reference.
• Figure 4: Overview of our proposed pipeline. The process begins with dataset preparation, where objects from ShapeNetCore.v2.PC15Kchang_ShapeNetInformationRich3D_2015 are aligned and mirrored along the $x = 0$ plane to obtain right-side half-objects ($x >= 0$), each containing 15,000 points. During generation, the synthesized half-objects are mirrored across the $x$-axis to reconstruct full shapes, which are subsequently evaluated using the original validation set.
• Figure 5: Symmetry computation using the proposed measurement protocol in the generated shapes from PVD, S-PVD, LION, and S-LION across the three ShapeNet classes. Ground-truth results from ShapeNet are provided for reference.