Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using Diffusion Models

TL;DR

This work presents Surf-D, a novel method for generating high-quality 3D shapes as Surfaces with arbitrary topologies using Diffusion models, and proposes a new pipeline that employs a point-based AutoEncoder to learn a compact and continuous latent space for accurately encoding UDF and support high-resolution mesh extraction.

Abstract

We present Surf-D, a novel method for generating high-quality 3D shapes as Surfaces with arbitrary topologies using Diffusion models. Previous methods explored shape generation with different representations and they suffer from limited topologies and poor geometry details. To generate high-quality surfaces of arbitrary topologies, we use the Unsigned Distance Field (UDF) as our surface representation to accommodate arbitrary topologies. Furthermore, we propose a new pipeline that employs a point-based AutoEncoder to learn a compact and continuous latent space for accurately encoding UDF and support high-resolution mesh extraction. We further show that our new pipeline significantly outperforms the prior approaches to learning the distance fields, such as the grid-based AutoEncoder, which is not scalable and incapable of learning accurate UDF. In addition, we adopt a curriculum learning strategy to efficiently embed various surfaces. With the pretrained shape latent space, we employ a latent diffusion model to acquire the distribution of various shapes. Extensive experiments are presented on using Surf-D for unconditional generation, category conditional generation, image conditional generation, and text-to-shape tasks. The experiments demonstrate the superior performance of Surf-D in shape generation across multiple modalities as conditions. Visit our project page at https://yzmblog.github.io/projects/SurfD/.

Figures (11)

• Figure 1: Surf-D achieves high-quality Surface generation for detailed geometry and various topology using a Diffusion model. It achieves SOTA performance in various shape generation tasks, including unconditional generation, category conditional generation, sketch conditional shape generation, single-view reconstruction, and text-guided shape generation.
• Figure 2: Comparison of learning discrete grid-based field cheng2023sdfusion and continuous field (ours) for UDFs (top) and corresponding gradient errors (bottom).The darker the color, the smaller the values.
• Figure 3: Framework of Surf-D. We first encode surface points into a latent code $z$ by our encoder $\mathcal{E}$. A curriculum scheduler helps to train our model in an easy-to-hard sample order. Then we train the diffusion model in our latent space and various conditions can be added by a task-specific encoder $\tau$. Finally, the sampled latent code $z$ will be decoded to a UDF field for mesh extraction by our UDF decoder $\mathcal{D}$.
• Figure 4: Samples from unconditional generation. Our method produces high-quality and diverse shapes. We also calculate their average CD for each object in the training set to confirm that our model is capable of producing unique shapes.
• Figure 5: Qualitative results of Category Conditional Generation. Our method generates different categories of detailed 3D shapes with high quality and diversity.