Datasets:

RenxingIntelligence
/

OpenVTON

dataset_info:
  features:
    - name: source
      dtype: image
    - name: mask
      dtype: image
    - name: target
      dtype: image
    - name: caption
      dtype: string
    - name: category
      dtype: string
  splits:
    - name: train
      num_examples: 89927
    - name: validation
      num_examples: 4989
    - name: test
      num_examples: 5009
license: cc-by-nc-4.0
task_categories:
  - image-to-image
tags:
  - virtual-try-on
  - fashion
  - clothing

OpenVTON

A large-scale virtual try-on dataset containing ~100K clothing image pairs with garment masks. You

Dataset Structure

Each sample contains:

source: Garment image (clothing item)
mask: Garment segmentation mask
target: Person wearing the garment (ground truth)
caption: Text description of the clothing
category: Clothing category (e.g., pants, jeans, shirt)

Splits

Split	Samples
Train	89,927
Validation	4,989
Test	5,009
Total	99,925

Usage

from datasets import load_dataset

dataset = load_dataset("RenxingIntelligence/OpenVTON")
sample = dataset["train"][0]
sample["source"].show()  # garment image
sample["mask"].show()    # segmentation mask
sample["target"].show()  # person wearing garment
print(sample["caption"])
print(sample["category"])

Benchmark and Paper

This dataset is part of OpenVTON-Bench, a large-scale benchmark designed for the systematic evaluation of controllable virtual try-on (VTON) models.

OpenVTON-Bench is introduced in our paper:

OpenVTON-Bench: A Large-Scale Benchmark for Controllable Virtual Try-On 📄 Paper: https://arxiv.org/abs/2601.22725 💻 Code: https://github.com/RenxingIntelligence/OpenVTON-Bench

OpenVTON-Bench provides a standardized evaluation protocol for modern diffusion-based and transformer-based virtual try-on systems, enabling fair and reproducible comparison across different architectures.

About OpenVTON-Bench

OpenVTON-Bench is a large-scale, high-resolution benchmark designed for the systematic evaluation of controllable virtual try-on models.

Unlike existing datasets and evaluation protocols that struggle with texture details and semantic consistency, OpenVTON-Bench provides:

🖼️ ~100K Image Pairs with resolutions up to 1536×1536, enabling evaluation of fine-grained texture generation.
🏷️ Fine-Grained Taxonomy covering 20 garment categories for balanced semantic evaluation.
📐 Multi-Level Automated Evaluation, including:
- Pixel fidelity
- Garment consistency
- Semantic realism

This benchmark enables fair, reproducible, and scalable comparison across modern virtual try-on systems.

Citation

If you use this dataset or the benchmark in your research, please cite:

@misc{li2026openvtonbenchlargescalehighresolutionbenchmark,
      title={OpenVTON-Bench: A Large-Scale High-Resolution Benchmark for Controllable Virtual Try-On Evaluation}, 
      author={Jin Li and Tao Chen and Shuai Jiang and Weijie Wang and Jingwen Luo and Chenhui Wu},
      year={2026},
      eprint={2601.22725},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2601.22725}, 
}