MagicGarment: Benchmarking Single-Image Standalone 3D Garment Reconstruction

Jia Chen; Ruyang Zhao; Jun Bai; Chao Huang; Saishang Zhong; Xinrong Hu

cs.CVOct 14, 2026

MagicGarment: Benchmarking Single-Image Standalone 3D Garment Reconstruction

Authors: Jia Chen, Ruyang Zhao, Jun Bai, Chao Huang, Saishang Zhong, Xinrong Hu

Organizations: School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan 430200, China · Engineering Research Center of Hubei Province for Clothing Information, Wuhan Textile University, Wuhan 430200, China · School of Computer Science, McGill University, Montreal, QC H3A 0E9, Canada

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/.

Figures & tables

Figure 1: Overview of MagicGarment. Left: representative garment assets and aligned multimodal annotations. Right: the single-image standalone 3D garment reconstruction task.

Figure 2: Overview of the garment-aware reference baseline built on pretrained Hunyuan3D. GSR performs input-level garment refinement, while FTR-LoRA provides model-level adaptation of the frozen Hunyuan3D-DiT backbone.

Figure 3: Representative 3D garment assets across the 17 composite types in MagicGarment.

Dataset	Avg. #Vertices	Avg. #Faces	#Types	Method	#Annotations
MagicGarment	375,817	665,488	17	Professional 3D artist modeling	9
SewFactory	8,813	16,834	13	Simulation	3
CLOTH3D	7,473	14,329	6	Simulation	3
DeepFashion3D	21,782	42,613	10	Real scans	4
GarmentCode v2	14,035	26,384	N/A	Simulation	4

Table 1: Comparison of representative 3D garment datasets.

Figure 4: Qualitative comparison on MagicGarment. Each example shows the input image, ground-truth mesh, and reconstructions from our garment-aware baseline and representative image-to-3D methods.

	Geometry		Appearance			Normal		Semantic
Method	CD $\downarrow$	F-score $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR-N $\uparrow$	LPIPS-N $\downarrow$	U1-I $\uparrow$	U1-T $\uparrow$	U2-I $\uparrow$	U2-T $\uparrow$	Uni-I $\uparrow$	Uni-T $\uparrow$
Hunyuan3D-2.1	4.2322	12.5148	17.6060	0.8193	0.2294	16.0559	0.2386	0.1139	0.0612	0.3975	0.2845	0.3411	0.2601
AIpparel	34.4063	5.0719	14.1433	0.7452	0.4192	9.5482	0.4608	0.0613	-0.0336	0.1276	0.0731	0.0837	0.0952
ChatGarment	39.4414	4.6322	14.1324	0.7617	0.4161	9.8198	0.4453	0.0607	-0.0324	0.1203	0.0696	0.0899	0.1046
CRM	11.5174	6.5935	18.5177	0.8252	0.2431	15.3413	0.2646	0.1253	0.0688	0.3078	0.2043	0.2712	0.1966
LGM	16.5124	5.2953	18.1696	0.8254	0.2531	15.5809	0.2594	0.1392	0.0599	0.2844	0.1855	0.2487	0.1987

Table 2: Zero-shot benchmark results on MagicGarment. U1/U2 denote ULIP-1/2, Uni denotes Uni3D, and I/T denote image/text similarity.

	Geometry		Appearance			Normal		Semantic
Method	CD $\downarrow$	F-score $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR-N $\uparrow$	LPIPS-N $\downarrow$	U1-I $\uparrow$	U1-T $\uparrow$	U2-I $\uparrow$	U2-T $\uparrow$	Uni-I $\uparrow$	Uni-T $\uparrow$
Hunyuan3D-2.1	2.7026	14.0527	19.0146	0.8380	0.1889	17.6048	0.2069	0.1162	0.0597	0.3900	0.2813	0.3475	0.2584
InstantMesh	11.3486	8.0174	17.6216	0.8475	0.1962	18.0185	0.2140	0.1402	0.0586	0.3751	0.2657	0.3198	0.2455
TRELLIS	4.3491	11.6548	17.9737	0.7390	0.2810	13.6614	0.2659	0.1227	0.0592	0.3629	0.2418	0.3094	0.2199
TRELLIS.2	3.2618	15.2347	17.8024	0.7824	0.2518	14.6542	0.2292	0.1334	0.0599	0.3588	0.2342	0.3035	0.2156
Wonder3D	11.0617	10.4935	16.0713	0.8296	0.2480	16.6064	0.2332	0.1359	0.0340	0.3642	0.2461	0.3008	0.2467

Table 3: In-domain adaptation results on MagicGarment. U1/U2 denote ULIP-1/2, Uni denotes Uni3D, and I/T denote image/text similarity.

3D Garment Reconstruction

See all 15 papers

Forwardrobe: Garment-Aware Gaussian Avatars from a Single Image
Jul 31, 2026Daisheng Jin, Shuyun Wang, Ying He3D Gaussian Splatting 3D Reconstruction
Nanyang Technological University · Singapore · The University of Queensland +1
Stitched Embeddings: A Unified Latent Space for 3D Garments and 2D Patterns
Jul 1, 2026Andrea Sanchietti, Riccardo Marin, Bharat Lal Bhatnagar +23D Reconstruction 3D Garment Reconstruction
University of Tübingen, Germany · Tübingen AI Center · Technical University of Munich, Germany +3
Spatio-Temporal Garment Reconstruction Using Diffusion Mapping via Pattern Coordinates
Feb 27, 2026Yingxuan You, Ren Li, Corentin Dumery +33D Garment Reconstruction 3D Reconstruction
CVLab, École Polytechnique Fédérale de Lausanne (EPFL), Switzerland · Mohamed bin Zayed University of Artificial Intelligence, United Arab Emirates · Pinscreen and Mohamed bin Zayed University of Artificial Intelligence, United Arab Emirates

Jul 31, 2026J/K move · Enter open · S save

Forwardrobe: Garment-Aware Gaussian Avatars from a Single Image

Daisheng Jin, Shuyun Wang, Ying He

Nanyang Technological University · Singapore · The University of Queensland+1

3D Gaussian Splatting 3D Reconstruction 3D Garment Reconstruction 3D Avatar Generation Human Avatar Animation 3D Human Reconstruction+2

Open paper AskPDF arXiv

Reconstructing animatable 3D human avatars from a single image remains particularly challenging for loose garments, whose geometry and motion cannot be adequately represented by body-aligned topology and skinning. We present Forwardrobe, a feed-forward framework for reconstructing garment-aware Gaussian avatars from a single image. Forwardrobe explicitly separates clothing from the body in canonical Gaussian space and equips the garment layer with continuity-aware geometry and skinning initialization, pose-conditioned non-rigid deformation, and appearance adaptation. These designs improve garment reconstruction and visual quality during animation, particularly for skirts and dresses. The separated garment layer additionally forms an independently controllable 3D asset, enabling garment editing, transfer, and 3D virtual try-on. Experiments demonstrate improved garment reconstruction quality and greater flexibility in garment manipulation compared with existing single-image avatar reconstruction methods.

Abstract

3D Garment Reconstruction

Explore similar work

Forwardrobe: Garment-Aware Gaussian Avatars from a Single Image

Stitched Embeddings: A Unified Latent Space for 3D Garments and 2D Patterns

Spatio-Temporal Garment Reconstruction Using Diffusion Mapping via Pattern Coordinates