C-VTON: Context-Driven Image-Based Virtual Try-On Network

[paper pdf]

Abstract

Image-based virtual try-on techniques have shown great promise for enhancing the user-experience and improving customer satisfaction on fashion-oriented e-commerce platforms. However, existing techniques are currently still limited in the quality of the try-on results they are able to produce from input images of diverse characteristics. In this work, we propose a Context-Driven Virtual Try-On Network (C-VTON) that addresses these limitations and convincingly transfers selected clothing items to the target subjects even under challenging pose configurations and in the presence of self-occlusions. At the core of the C-VTON pipeline are: (i) a geometric matching procedure that efficiently aligns the target clothing with the pose of the person in the input images, and (ii) a powerful image generator that utilizes various types of contextual information when synthesizing the final try-on result. C-VTON is evaluated in rigorous experiments on the VITON and MPV datasets and in comparison to state-of-the-art techniques from the literature. Experimental results show that the proposed approach is able to produce photo-realistic and visually convincing results and significantly improves on the existing state-of-the-art.

Requirements

• Python (3.8.10)
• opencv-python (4.5.1.48)
• torch (1.7.1+cu110)
• torchvision (0.8.2+cu110)
• lpips (0.1.4)
• matplotlib (3.3.4)
• pandas (1.2.0)

The experiments were conducted on one (for image resolutions 256×192) or two (for image resolution 512×368) NVIDIA RTX 3090 GPUs.

Testing

To avoid setting up this repository only to obtain results presented in the paper, please see the generated test results here (for VITON), here (for VITON-HD) and here (for MPV). You can find respective test splits alongside provided preprocessed data here (VITON) and here (MPV).

Running tests yourself requires some setting up to do:

• Install requirements.
• For VITON / VITON-HD:
  • Download our preprocessed data and extract them to folder ./data/. Since we are not allowed to share the actual dataset, contact the authors of the original VITON paper, ask them for the full-resolution dataset and download it to appropriate folders.
• For MPV:
  • Download our preprocessed data and extract them to folder ./data/. Additionally, obtain the version of the MPV dataset in resolution 256×192 and extract it alongside preprocessed data.
• Download pretrained BPGM model for VITON / VITON-HD or MPV and put it in ./bpgm/checkpoints.
• Download pretrained C-VTON model for VITON, VITON-HD or MPV and put it in ./checkpoints.
• Run test scripts:

cd C-VTON
./scripts/test_{dataset}.sh

Make sure you use our test split to obtain the same results.

Training

To train the models from scratch, download all data mentioned in previous section and install requirements. Then, similarly to testing, run appropriate scripts from the scripts folder:

cd C-VTON
./scripts/train_{dataset}.sh

When training using VITON-HD dataset make sure to delete the second line in train.py in order to use all available GPUs.

Citation

If you use code or results from this repository, please cite the following publication:

@InProceedings{Fele_2022_WACV,
    author = {Fele, Benjamin and Lampe, Ajda and Peer, Peter and Struc, Vitomir},
    title = {C-VTON: Context-Driven Image-Based Virtual Try-On Network},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month = {January},
    year = {2022}
}

Parts of this repository are based on implementation of You Only Need Adversarial Supervision for Semantic Image Synthesis.

Acknowledgements

Supported in parts by the Slovenian Research Agency ARRS through the Research Programmes P2-0250(B) "Metrology and Biometric System" and P2--0214 (A) “Computer Vision”, the ARRS Project J2-2501(A) "DeepBeauty" and the ARRS junior researcher program.