FIT: A Large-Scale Dataset for Fit-Aware Virtual Try-On

2026-04-09 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, quick

Summary

A new large-scale dataset, FIT (Fit-Inclusive Try-on), has been introduced to address the critical gap in virtual try-on (VTO) methods regarding accurate garment fit. Current VTO systems often fail to depict how garments of varying sizes look on different body types, defaulting to well-fitted results due to a lack of precise garment and body size data, especially for "ill-fit" scenarios. The FIT dataset comprises over 1.13 million try-on image triplets, each accompanied by exact body and garment measurements. Its creation involved a scalable synthetic strategy: programmatically generating 3D garments with GarmentCode and using physics simulations for realistic draping, employing a re-texturing framework to convert synthetic renderings into photorealistic images while maintaining geometry, and integrating person identity preservation for supervised training. This dataset enables the development of fit-aware VTO models and establishes a new benchmark for future research.

Key takeaway

For research scientists developing virtual try-on systems, you should integrate the FIT dataset to improve the accuracy of garment fit representation. This dataset directly addresses the limitation of current VTO methods that overlook "ill-fit" cases, allowing your models to generate more realistic and size-accurate try-on images. Consider leveraging its structured measurements to train models capable of distinguishing and depicting various garment-to-body size relationships.

Key insights

The FIT dataset enables fit-aware virtual try-on by providing precise garment and body size information.

Principles

• Realistic garment fit requires precise size data.
• Synthetic data generation can overcome collection challenges.

Method

The FIT dataset is generated by programmatically creating 3D garments, simulating physics for draping, re-texturing synthetic renderings into photorealistic images, and preserving person identity for paired training.

In practice

• Use GarmentCode for 3D garment generation.
• Employ physics simulation for realistic draping.
• Apply re-texturing to create photorealistic synthetic data.

Topics

• Virtual Try-On
• Garment Fit
• Large-Scale Dataset
• Synthetic Data Generation
• GarmentCode

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• 🪞1.1M Metric VTON Dataset🪞 👉Google's Fit-Inclusive Try-on: large-scale VTO dataset... — The FIT dataset offers over 1.13M image triplets with precise measurements for improved virtual try-on.
• ETCH-X: Robustify Expressive Body Fitting to Clothed Humans with Composable Datasets — ETCH-X robustly fits expressive body models to clothed 3D scans using a modular, tightness-aware, dense correspondence approach.
• 🔥Vanast: VTON w/ Human Animation🔥 👉SNU unveils a novel unified framework that generates... — Vanast generates human animation videos with garment transfer from single images and pose guidance.
• A new way to create a digital wardrobe from your Google Photos — AI-powered digital wardrobe creation in Google Photos streamlines outfit planning and virtual try-on.
• Efficiently Linking Real Scenes with Synthetic Data Generation for AI-based Cognitive Robotics and Computer Vision Applications — Bridging the sim-to-real domain gap via linked synthetic and real data is crucial for scalable AI robotics.
• SQuadGen: Generating Simple Quad Layouts via Chart Distance Fields — SQuadGen uses diffusion models and Chart Distance Fields to generate simple, artist-friendly quad layouts for 3D shapes.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.