New Story

Wonder3D: Evaluating The Quality of The Reconstructed Geometry of Different Methods

by RingiJanuary 3rd, 2025

Too Long; Didn't Read

We evaluate the quality of the reconstructed geometry of different methods. The quantitative results are summarized in Table 1

featured image - Wonder3D: Evaluating The Quality of The Reconstructed Geometry of Different Methods

‘colorful geometry’ Image created by HackerNoon AI Image Generator

Table of Links

Abstract and 1 Introduction

2. Related Works

2.1. 2D Diffusion Models for 3D Generation

2.2. 3D Generative Models and 2.3. Multi-view Diffusion Models

3. Problem Formulation

3.1. Diffusion Models

3.2. The Distribution of 3D Assets

4. Method and 4.1. Consistent Multi-view Generation

4.2. Cross-Domain Diffusion

4.3. Textured Mesh Extraction

5. Experiments

5.1. Implementation Details

5.2. Baselines

5.3. Evaluation Protocol

5.4. Single View Reconstruction

5.5. Novel View Synthesis and 5.6. Discussions

6. Conclusions and Future Works, Acknowledgements and References

5.4. Single View Reconstruction

We evaluate the quality of the reconstructed geometry of different methods. The quantitative results are summarized in Table 1, and the qualitative comparisons are presented in Fig. 6. Shap-E [25] tends to produce incomplete and distorted meshes. SyncDreamer [33] generates shapes that are roughly aligned with the input image but lack detailed geometries, and the texture quality is subpar. One-2-3- 45 [30] attempts to reconstruct meshes from the multiviewinconsistent outputs of Zero123 [31]. While it can capture coarse geometries, it loses important details in the process. In comparison, our method stands out by achieving the highest reconstruction quality, both in terms of geometry and textures.