Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders

TL;DR

Dora-VAE introduces a sharp edge sampling and dual attention mechanism to improve 3D shape reconstruction in VAEs, focusing on complex geometric regions, and proposes Dora-bench for evaluating shape complexity and reconstruction quality.

Contribution

The paper presents Dora-VAE with a novel sampling and attention strategy, and introduces Dora-bench for systematic evaluation of 3D shape VAE reconstructions.

Findings

Dora-VAE achieves comparable reconstruction quality to state-of-the-art methods.

Dora-VAE uses at least 8 times fewer latent codes than dense models.

Dora-bench effectively quantifies shape complexity and reconstruction accuracy.

Abstract

Recent 3D content generation pipelines commonly employ Variational Autoencoders (VAEs) to encode shapes into compact latent representations for diffusion-based generation. However, the widely adopted uniform point sampling strategy in Shape VAE training often leads to a significant loss of geometric details, limiting the quality of shape reconstruction and downstream generation tasks. We present Dora-VAE, a novel approach that enhances VAE reconstruction through our proposed sharp edge sampling strategy and a dual cross-attention mechanism. By identifying and prioritizing regions with high geometric complexity during training, our method significantly improves the preservation of fine-grained shape features. Such sampling strategy and the dual attention mechanism enable the VAE to focus on crucial geometric details that are typically missed by uniform sampling approaches. To systematically evaluate VAE reconstruction quality, we additionally propose Dora-bench, a benchmark that quantifies shape complexity through the density of sharp edges, introducing a new metric focused on reconstruction accuracy at these salient geometric features. Extensive experiments on the Dora-bench demonstrate that Dora-VAE achieves comparable reconstruction quality to the state-of-the-art dense XCube-VAE while requiring a latent space at least 8$\times$ smaller (1,280 vs. > 10,000 codes).