MagicGarment provides new benchmark for 3D garment modeling
MagicGarment: Benchmarking Single-Image Standalone 3D Garment Reconstruction
Computer Vision and Pattern Recognition
Summary
3D modeling of clothes from just one photo is hard because garments have thin shapes, open edges, and layers. The authors created MagicGarment, a large collection of detailed 3D garment models with matching images and labels to test and compare reconstruction methods. They also improved an existing 3D model by adding garment-specific features to make it understand clothes better. Their tests show that this new benchmark and adaptation help improve 3D clothing reconstruction from single images.
What this means in practice
- •For fashion designers: Test and improve 3D clothing models created from single photos to speed up virtual garment prototyping.
- •For game developers: Use the benchmark and adapted models to generate realistic 3D clothes from images for virtual characters and avatars.
Authors
Jia Chen, Ruyang Zhao, Jun Bai, Chao Huang, Saishang Zhong, Xinrong Hu
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. Project page: https://1019538157.github.io/MagicGarment/.