GarVerseLOD: High‑Fidelity 3D Garment Reconstruction from a Single In‑the‑Wild Image using a Dataset with Levels of Details

Dataset Overview

• Title: GarVerseLOD: High‑Fidelity 3D Garment Reconstruction from a Single In‑the‑Wild Image using a Dataset with Levels of Details
• Authors:
  • Zhongjin Luo¹
  • Haolin Liu^2,1
  • Chenghong Li^2,1
  • Wanghao Du²
  • Zirong Jin²
  • Wanhu Sun²
  • Yinyu Nie³
  • Weikai Chen⁴
  • Xiaoguang Han^#1,2
• Institutions:
  1. SSE, CUHKSZ
  2. FNii, CUHKSZ
  3. Huawei Noah’s Ark Lab
  4. DCC Algorithm Research Center, Tencent Games
• Publication: ACM Transactions on Graphics (SIGGRAPH Asia 2024)

Dataset Resources

• ARXIV: https://arxiv.org/abs/2411.03047
• PDF: https://garverselod.github.io/GarVerseLOD.pdf
• CODE: https://github.com/zhongjinluo/GarVerseLOD
• DATA: https://github.com/zhongjinluo/GarVerseLOD

Dataset Summary

• Abstract: We propose a hierarchical framework that leverages garment shape and deformation priors from the GarVerseLOD dataset to recover garment details at multiple levels. Given a single RGB image of a clothed person captured from the web, our method generates high‑fidelity 3D garment meshes that exhibit realistic deformations and align well with the input image.

Detailed Description

• Method Overview: From an input RGB image, the pipeline first estimates a T‑shaped garment shape and uses a predicted SMPL body to compute pose‑dependent deformations. A pixel‑aligned network then reconstructs fine implicit garment details, while a geometry‑aware boundary estimator predicts garment silhouettes. Finally, a registration step produces a topologically consistent mesh with open boundaries.
• Dataset Characteristics: GarVerseLOD includes 6,000 high‑quality garment models manually crafted by professional artists, featuring fine‑grained geometry. The dataset is structured with three LODs, ranging from coarse stylized shapes to pose‑blended garments with pixel‑aligned details.
• Evaluation: Extensive experiments on numerous in‑the‑wild images demonstrate that GarVerseLOD enables generation of garment components that surpass existing methods in quality and robustness to variations in pose, illumination, occlusion, and deformation.

Citation

@article{luo2024garverselod,
  title={GarVerseLOD: High‑Fidelity 3D Garment Reconstruction from a Single In‑the‑Wild Image using a Dataset with Levels of Details},
  author={Luo, Zhongjin and Liu, Haolin and Li, Chenghong and Du, Wanghao and Jin, Zirong and Nie, Yinyu and Chen, Weikai and Han, Xiaoguang},
  journal={ACM Transactions on Graphics (TOG)},
  year={2024}
}