MagicGarment: Benchmarking Single-Image Standalone 3D Garment Reconstruction

Jia Chen1,2†, Ruyang Zhao1†, Jun Bai3*, Chao Huang1, Saishang Zhong1, Xinrong Hu1
1School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan 430200, China 2Engineering Research Center of Hubei Province for Clothing Information, Wuhan Textile University, Wuhan 430200, China 3School of Computer Science, McGill University, Montreal, QC H3A 0E9, Canada
*Corresponding author. †Equal contribution.
Paper Code Dataset

Abstract

Single-image 3D reconstruction has advanced rapidly, yet standalone garments remain challenging due to thin-shell geometry, open boundaries, and layered structures. Existing datasets and evaluation protocols provide limited support for systematic evaluation in this setting. We introduce MagicGarment, a benchmark for single-image standalone 3D garment reconstruction with 1,518 professionally constructed assets across 17 composite garment types, aligned multimodal annotations, and a unified evaluation protocol. We further establish a garment-aware reference baseline by adapting a pretrained image-to-3D backbone at two levels: Grounded Semantic Refinement (GSR) provides garment-focused visual conditioning, while Flow-Time-Route LoRA (FTR-LoRA) adapts the frozen 3D generative backbone using flow-timestep, token-routing, and expert-load information. Experiments demonstrate MagicGarment's benchmark utility and the effectiveness of garment-aware adaptation.

Overview of MagicGarment

Method

MagicGarment network structure diagram

Dataset

Representative garment assets in MagicGarment

More Results

MagicGarment qualitative experimental diagram

Interactable Meshes

Hanfu

Long-Sleeve Dress

Three-Piece Set

Citation

@misc{chen2026magicgarmentbenchmarkingsingleimagestandalone,
  title={MagicGarment: Benchmarking Single-Image Standalone 3D Garment Reconstruction},
  author={Jia Chen and Ruyang Zhao and Jun Bai and Chao Huang and Saishang Zhong and Xinrong Hu},
  year={2026},
  eprint={2610.19623},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2610.19623}
}