Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GarmentGS: Point-Cloud Guided Gaussian Splatting for High-Fidelity Non-Watertight 3D Garment Reconstruction

Zhihao Tang, Shenghao Yang, Hongtao Zhang, Mingbo Zhao

TL;DR

GarmentGS introduces a dense point-cloud–guided Gaussian splatting pipeline to enable high-fidelity, non-watertight 3D garment reconstruction from multi-view images. By generating a dense point cloud with fast MVS, moving and reorienting 3D Gaussian primitives, flattening them into 2D disks, and aligning their normals to the point cloud, the method achieves accurate surface geometry while producing single-layer meshes suitable for fabric simulation. A mesh-denoising step using LOF ensures removal of internal fragments, resulting in clean garments ideal for virtual try-on and downstream workflows. Evaluations on DeepFashion3D-v2 demonstrate superior rendering quality and geometric accuracy compared with state-of-the-art Gaussian-based methods, while significantly reducing reconstruction time to roughly 10 minutes for dense point clouds and enabling real-time-like rendering.

Abstract

Traditional 3D garment creation requires extensive manual operations, resulting in time and labor costs. Recently, 3D Gaussian Splatting has achieved breakthrough progress in 3D scene reconstruction and rendering, attracting widespread attention and opening new pathways for 3D garment reconstruction. However, due to the unstructured and irregular nature of Gaussian primitives, it is difficult to reconstruct high-fidelity, non-watertight 3D garments. In this paper, we present GarmentGS, a dense point cloud-guided method that can reconstruct high-fidelity garment surfaces with high geometric accuracy and generate non-watertight, single-layer meshes. Our method introduces a fast dense point cloud reconstruction module that can complete garment point cloud reconstruction in 10 minutes, compared to traditional methods that require several hours. Furthermore, we use dense point clouds to guide the movement, flattening, and rotation of Gaussian primitives, enabling better distribution on the garment surface to achieve superior rendering effects and geometric accuracy. Through numerical and visual comparisons, our method achieves fast training and real-time rendering while maintaining competitive quality.

GarmentGS: Point-Cloud Guided Gaussian Splatting for High-Fidelity Non-Watertight 3D Garment Reconstruction

TL;DR

GarmentGS introduces a dense point-cloud–guided Gaussian splatting pipeline to enable high-fidelity, non-watertight 3D garment reconstruction from multi-view images. By generating a dense point cloud with fast MVS, moving and reorienting 3D Gaussian primitives, flattening them into 2D disks, and aligning their normals to the point cloud, the method achieves accurate surface geometry while producing single-layer meshes suitable for fabric simulation. A mesh-denoising step using LOF ensures removal of internal fragments, resulting in clean garments ideal for virtual try-on and downstream workflows. Evaluations on DeepFashion3D-v2 demonstrate superior rendering quality and geometric accuracy compared with state-of-the-art Gaussian-based methods, while significantly reducing reconstruction time to roughly 10 minutes for dense point clouds and enabling real-time-like rendering.

Abstract

Traditional 3D garment creation requires extensive manual operations, resulting in time and labor costs. Recently, 3D Gaussian Splatting has achieved breakthrough progress in 3D scene reconstruction and rendering, attracting widespread attention and opening new pathways for 3D garment reconstruction. However, due to the unstructured and irregular nature of Gaussian primitives, it is difficult to reconstruct high-fidelity, non-watertight 3D garments. In this paper, we present GarmentGS, a dense point cloud-guided method that can reconstruct high-fidelity garment surfaces with high geometric accuracy and generate non-watertight, single-layer meshes. Our method introduces a fast dense point cloud reconstruction module that can complete garment point cloud reconstruction in 10 minutes, compared to traditional methods that require several hours. Furthermore, we use dense point clouds to guide the movement, flattening, and rotation of Gaussian primitives, enabling better distribution on the garment surface to achieve superior rendering effects and geometric accuracy. Through numerical and visual comparisons, our method achieves fast training and real-time rendering while maintaining competitive quality.
Paper Structure (14 sections, 7 equations, 4 figures, 2 tables)

This paper contains 14 sections, 7 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-View Stereo
  4. Surface Reconstruction with 3D Gaussians
  5. Method
  6. Preliminary
  7. Gaussian Optimization
  8. Mesh Denoising
  9. Experiments
  10. Experiment Settings
  11. Comparisons
  12. Ablations
  13. Conclusion
  14. Acknowledgments

Figures (4)

  • Figure 1: Our method reconstructs 3D garments from multi-view images through fast dense point cloud reconstruction and Gaussian primitive optimization, achieving high-fidelity rendering effect while generating non-watertight meshes that preserve complex topological.
  • Figure 2: Overview of our method. We first (a) pull scattered 3D Gaussian primitives towards their nearest points in the dense point cloud, while (b) flattening 3D Gaussian ellipsoids into 2D Gaussian elliptical disks, then (c) rotate the 2D Gaussian elliptical disks at corresponding positions according to the normal directions of points in the dense point cloud to align the Gaussian primitives precisely with the clothing surface.
  • Figure 3: Comparison between ours and other methods on multi-view reconstruction on DeepFashion3D dataset. Top row: rendering effect. Bottom 3 rows: reconstructed meshes.
  • Figure 4: Results for some garments. First column: ground truth. Second column: rendering effect. Remaining 5 columns: reconstructed meshes.