MV-Fashion/MV-Fashion
Updated • 311 • 1
AI & ML interests
Technologies for digital fashion.
Recent Activity
View all activity
Organization Card
MV-Fashion: Towards Enabling Virtual Try-On and Size Estimation with Multi-View Paired Data
MV-Fashion is a large-scale, multi-view video dataset engineered for fashion-specific computer vision research, introduced at CVPR 2026 (Highlight) by the Computer Vision Center (Barcelona).
📄 Paper · 🌐 Project Page
Abstract
Existing 4D human datasets often fall short for fashion-specific research, lacking either realistic garment dynamics or task-specific annotations. Synthetic datasets suffer from a realism gap, whereas real-world captures lack the detailed annotations and paired data required for virtual try-on (VTON) and size estimation tasks.
To bridge this gap, we introduce MV-Fashion, a massive multi-view video dataset engineered for domain-specific fashion analysis. MV-Fashion captures complex, real-world garment dynamics across 80 diverse subjects wearing multiple layered outfits. Crucially for VTON applications, it provides paired data: synchronized multi-view captures of worn garments alongside their corresponding flat, catalogue images.
Key Features
👥 Diverse Subjects
80 subjects spanning a wide range of demographics:
Gender: Male (50.6%), Female (45.7%), Non-binary (3.7%)
Age: 18–24 (53%), 25–34 (35%), 35–44 (10%), 45–54 (2%)
BMI: <18.5 (15%), 18.5–24.9 (69%), 25–29.9 (13%), 30+ (4%)
👕 Garment Coverage
754 garments across 14 categories, including single, double, and triple-layer outfits:
| Category | Share |
|---|---|
| Shirt / Blouse | 38.3% |
| Pants | 21.2% |
| Shorts | 11.7% |
| Dress | 5.2% |
| Jacket | 5.1% |
| Jeans | 4.8% |
| Skirt | 4.5% |
| Sweater | 3.2% |
| Other | 6.0% |
Fit distribution: Slim (8.1%), Regular (64.1%), Loose (24.7%)
📷 Multi-View Capture System
| Component | Count | Notes |
|---|---|---|
| Total synchronized cameras | 68 | Controlled studio environment |
| RGB global-shutter cameras | 60 | High-fidelity appearance capture |
| Depth / 4K cameras | 8 | Geometry and fine-detail capture |
Full 360° coverage enabling novel view synthesis.
🔄 Paired Data for Virtual Try-On
Each sequence includes:
- Multi-view synchronized video of the worn garment
- Corresponding flat catalogue image of the same garment
- Enabling direct supervision for image-based VTON models
🏷️ Rich Annotations
| Annotation | Description |
|---|---|
| SMPL-X body fits | Parametric 3D body shape and pose |
| 3D point clouds | Dense geometry per frame |
| Segmentation masks | Pixel-level garment and body part labels |
| Text descriptions | Natural language garment attributes |
| Material elasticity | 5-level fabric stretch classification |
| Garment fit labels | Slim / regular / loose per garment |
Elasticity distribution: Inelastic (37.9%), Level 2 (29.2%), Level 3 (25.1%), Level 4 (6.2%), Highly Elastic (0.5%)
💃 Challenging Garment Dynamics
- Multi-layer outfits (single, double, triple layers)
- Varied styling: tucked, rolled, layered combinations
- Natural motion sequences for realistic deformation capture
- Diverse poses and body movements per sequence
Intended Use Cases
| Use Case | Description |
|---|---|
| Virtual Try-On (VTON) | Image- and video-based garment transfer |
| Clothing Size Estimation | Body measurement and fit prediction |
| Novel View Synthesis | 3D garment reconstruction from multi-view input |
| Human Body Reconstruction | 4D pose and shape estimation |
| Garment Segmentation | Pixel-level clothing parsing |
| Fashion Analysis | Attribute recognition, style classification |
License
MV-Fashion is released for non-commercial academic research use only. By downloading or using this dataset, you agree to the terms of the MV-Fashion Dataset Agreement.
Citation
If you use MV-Fashion in your research, please cite:
@InProceedings{Laczko_2026_CVPR,
author = {Laczk\'o, Hunor and Jia, Libang and Truong, Loc-Phat and Hern\'andez, Diego and Escalera, Sergio and Gonzalez, Jordi and Madadi, Meysam},
title = {MV-Fashion: Towards Enabling Virtual Try-On and Size Estimation with Multi-View Paired Data},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2026},
pages = {42810-42823}
}
Team
models 0
None public yet