You can deploy this model and start earning money today!

NguyenDinhHieu/
VITON-Extends

Name: VITON-Extends
Availability: InStock

NguyenDinhHieu/VITON-Extends-DB

mit

Views

No views yet

VITON-Extends — Model weights & inference bundle

Enhancing Pose Adaptability in Virtual Try-On Systems

Author	Affiliation	ORCID
Nguyen Dinh Hieu	FPT University, Hanoi, Vietnam	0009-0002-6683-8036
Tran Minh Khuong	FPT University, Hanoi, Vietnam	—
Phan Duy Hung	FPT University, Hanoi, Vietnam	0000-0002-6033-6484

Contact: hieundhe180318@fpt.edu.vn, khuongtmhe180089@fpt.edu.vn, hungpd2@fe.edu.vn

What is in this Hub repository?

This model repo ships two top-level folders:

Folder on Hub	Role
`VITON-Extends_test/`	Code and assets to run inference / demo (`test.py`, networks, options, etc.).
`VITON-Extends-Train/`	Training-side bundle (scripts, configs, or checkpoints—whatever you packaged locally under that directory).

Download the full repo (or a subset with allow_patterns) and point your local paths to these folders as in the GitHub README.

Dataset (images) is hosted separately: NguyenDinhHieu/VITON-Extends-DB (see that dataset card for Train.zip / Test.zip and extraction).

Abstract

Garment fitting in virtual try-on often fails under complex poses, occlusions, and misalignment between person and garment. VITON-Extends improves pose adaptability and garment warping with a global appearance flow model, StyleGAN-style global modulation, and a local flow refinement stage. On the VITON benchmark, results are strong especially in challenging poses.

Paper: Springer LNCS (IUKM 2025), DOI 10.1007/978-981-96-4606-7_21
Code: github.com/nguyendinhhieu1309/VITON-Extends

Quick download (Python)

python

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="NguyenDinhHieu/VITON-Extends",
    local_dir="./VITON-Extends_hf",
)
# Then use ./VITON-Extends_hf/VITON-Extends_test/ and ./VITON-Extends_hf/VITON-Extends-Train/

Environment (reference)

Versions below match the paper / reference setup; your local VITON-Extends_* trees may ship their own requirements.txt—prefer those for exact pins.

Component	Reference version
PyTorch	2.2.1+cu118 (example)
TorchVision	0.17.1+cu118
CuPy	13.3.0
OpenCV	4.10.0
Python	3.12 (or as in project env)

Training & testing (outline)

Data: Use VITON-Extends-DB — unzip Train.zip / Test.zip, set dataroot to train/ or test/ as in the dataset card.
Checkpoints: Place warping / generation weights where the GitHub repo expects (e.g. under checkpoints/VITON-Extends/).
Train: Run the shell scripts under scripts/ from the GitHub repository (parser-based then parser-free stages).
Test: From the downloaded VITON-Extends_test/ tree, follow repo instructions, e.g.
python test.py --name demo --resize_or_crop None --batchSize 1 --gpu_ids 0

For FID and extra assets, see links in the GitHub README.

Results (qualitative)

Citation

bibtex

@inproceedings{hieu2025vitonextends,
  title     = {Enhancing Pose Adaptability in Virtual Try-On Systems},
  author    = {Hieu, Nguyen Dinh and Khuong, Tran Minh and Hung, Phan Duy},
  booktitle = {Integrated Uncertainty in Knowledge Modelling and Decision Making (IUKM 2025)},
  series    = {Lecture Notes in Computer Science},
  volume    = {15585},
  publisher = {Springer},
  address   = {Singapore},
  year      = {2025},
  doi       = {10.1007/978-981-96-4606-7_21}
}

Acknowledgements

Built on virtual try-on and flow-based clothed-person generation ideas; base code lineage includes ClothFlow. Full credits appear in the GitHub repository.