TryOffDiff

The official repository of the papers:

1. "TryOffDiff: Virtual-Try-Off via High-Fidelity Garment Reconstruction using Diffusion Models",
2. "MGT: Extending Virtual Try-Off to Multi-Garment Scenarios"

TL;DR:

---

🎉 News

• [2025-07-25]: Paper1 accepted at BMVC'25, will be presented on Nov 24-27, 2025 (tbd).
• [2025-07-11]: Paper2 accepted at ICCV'25 Workshop, will be presented on Oct 20th, 2025.
• [2025-07-10]: Code for new features made available.
• [2025-04-17]: Paper2 (follow-up work) appeared on arXiv with improvements, e.g. multi-garment try-off.
• [2025-03-26]: Demo is accepted at CVPR'25 Demo Track, presented on June 13, 2025.
• [2024-12-03]: Training, Inference, and Evaluation scripts made available.
• [2024-11-27]: Paper1 appeared on arXiv.

---

Usage

Please refer to the instructions.

---

Project Organization

The following project/directory structure is adopted: Cookiecutter Data Science-v2 by DrivenData

├── notebooks/           <- Jupyter notebooks
├── references/          <- Manuals and all other explanatory materials.
├── LICENSE
├── README.md
├── pyproject.toml       <- Project configuration file with package metadata
|
└── tryoffdiff/          <- Source code for use in this project.
    ├── modeling/
    │   ├── __init__.py
    │   ├── eval.py      <- Code to evaluate models
    │   ├── model.py     <- Model implementations
    │   ├── predict.py   <- Code to run model inference with trained models
    │   └── train.py     <- Code to train models
    |
    ├── __init__.py      <- Makes `tryoffdiff` a Python module
    ├── config.py        <- Store configuration variables for training and inference
    ├── dataset.py       <- Download and clean datasets VITON-HD & Dress Code
    ├── features.py      <- Code to create features for modeling
    └── plots.py         <- Code to create visualizations

Acknowledgements

Our code relies on PyTorch, with 🤗 Diffusers for diffusion model components and 🤗 Accelerate for multi-GPU training.
We adopt Stable Diffusion-v1.4 as the base model and use SigLIP as the image encoder.
For evaluation, we use IQA_PyTorch, clean-fid, and DISTS-pytorch.

License

TL;DR: Not available for commercial use, unless the FULL source code is open-sourced!
This project is intended solely for academic research. No commercial benefits are derived from it.
The code, datasets, and models are published under the Server Side Public License (SSPL).

Citation

If you find this repository useful in your research, please consider giving a star ⭐ and a citation:

@inproceedings{velioglu2025tryoffdiff,
  title     = {TryOffDiff: Virtual-Try-Off via High-Fidelity Garment Reconstruction using Diffusion Models},
  author    = {Velioglu, Riza and Bevandic, Petra and Chan, Robin and Hammer, Barbara},
  booktitle = {BMVC},
  year      = {2025},
  note      = {\url{https://doi.org/nt3n}}
}
@inproceedings{velioglu2025mgt,
  title     = {MGT: Extending Virtual Try-Off to Multi-Garment Scenarios},
  author    = {Velioglu, Riza and Bevandic, Petra and Chan, Robin and Hammer, Barbara},
  booktitle = {ICCVW},
  year      = {2025},
  note      = {\url{https://doi.org/pn67}}
}