FitControler

TODO

• Release the Fit4Men dataset (publicly available through Google Drive)
• Release the FitControler training and evaluation code
• Release pretrained FitControler model weights

Fit4Men Dataset

Fit4Men is a male fashion dataset used by the FitControler project for fit-aware virtual try-on research. The dataset is now publicly available for download. It contains two garment categories, upper-body garments (upper) and lower-body garments (lower), together with training and test splits and garment-fit labels.

Download

• Google Drive: https://drive.google.com/file/d/11yDFh6JYcstnBrdbttsJBXl0tWNsgdMI/view?usp=sharing

After downloading, extract the archive to obtain the Fit4Men/ directory and preserve the directory structure shown below. The uncompressed dataset is approximately 10.94 GB and contains 85,778 files.

Dataset Structure

Fit4Men/
├── lower/
│   ├── train/
│   │   ├── cloth/
│   │   ├── densepose/
│   │   ├── image/
│   │   ├── mask/
│   │   ├── person_keypoints/
│   │   ├── person_mask/
│   │   └── seg/
│   ├── test/
│   │   ├── cloth/
│   │   ├── cloth_mask/
│   │   ├── densepose/
│   │   ├── image/
│   │   ├── mask/
│   │   ├── person_keypoints/
│   │   ├── person_mask/
│   │   └── seg/
│   ├── train.json
│   ├── test.json
│   └── test_unpairs.json
└── upper/
    ├── train/
    │   ├── cloth/
    │   ├── densepose/
    │   ├── image/
    │   ├── mask/
    │   ├── person_keypoints/
    │   ├── person_mask/
    │   └── seg/
    ├── test/
    │   ├── cloth/
    │   ├── cloth_mask/
    │   ├── densepose/
    │   ├── image/
    │   ├── mask/
    │   ├── person_keypoints/
    │   ├── person_mask/
    │   └── seg/
    ├── train.json
    ├── test.json
    └── test_unpairs.json

Dataset Statistics

Category	Training records	Test records	Approximate size	type labels
lower	2,174	544	7.09 GB	straight, tapered
upper	1,853	467	3.85 GB	loose, regular, slim

The complete dataset contains 85,077 PNG files, 695 JPG files, and 6 JSON files.

File Description

Directory/File	Description
cloth/	Garment images
image/	Model images
cloth_mask/	Garment masks, provided for the test split
mask/	Human-related masks
person_mask/	Human-region masks
person_keypoints/	Human keypoint annotations or visualizations
densepose/	DensePose results
seg/	Semantic segmentation results
train.json	Training manifest
test.json	Test manifest
test_unpairs.json	Unpaired test manifest

JSON Annotation Format

Each record contains a garment filename, a list of corresponding model-image filenames, and a garment-fit label:

{
  "cloth": "337.png",
  "image": ["692.png", "693.png", "694.png"],
  "type": "tapered"
}

• cloth: The garment-image filename.
• image: A list of model-image filenames corresponding to the garment. The list usually contains multiple elements, allowing one garment to be associated with a group of model images.
• type: The garment-fit category. The lower subset uses straight and tapered, while the upper subset uses loose, regular, and slim.

Usage

1. Download Fit4Men.zip from the Google Drive link above.
2. Extract the archive and place the resulting Fit4Men/ directory under the data root configured by your data loader.
3. Select either upper or lower, then load train.json, test.json, or test_unpairs.json as required by your task.

License

The Fit4Men dataset is released under the Fit4Men Dataset Non-Commercial License (F4M-NC) v1.0. It may be used only for non-commercial purposes such as academic research, teaching, evaluation, and personal study. Commercial use is prohibited. See LICENSE for the complete terms. Please contact the project maintainers for separate written permission before any commercial use.

Images, annotations, or other materials in the dataset may also be subject to rights held by their original owners. This license does not replace those rights or grant permissions beyond those that the dataset authors are legally able to provide.

Citation

If FitControler or the Fit4Men dataset contributes to your research, please cite the ECCV 2026 paper:

@inproceedings{yang2026fitcontroler,
  title     = {FitControler: Toward Fit-Aware Virtual Try-On},
  author    = {Yang, Lu and Liu, Yicheng and Zhou, Letian and Li, Yanan and Bai, Xiang and Lu, Hao},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}