zijiyingcai/ViT-HD

ViT-HD： the large-scale, high-definition dataset for video virtual try-on

Paper

Updates

• 2025/12/24: The large-scale ViT-HD dataset is available.
• 2025/12/24: Our KeyTailor paper is available.

Key Features

Dataset Comparison. We compare our ViT-HD with existing video virtual try-on datasets along four dimensions: resolution, garment diversity (multi-class), video content quality (no start-frame overexposure and intact subject integrity), and data scale.

Our proposed ViT-HD contains 15, 070 samples featuring 178 diverse garment styles, each with a resolution of 810×1080.

Dataset overview

An overview of ViT-HD is as follows:

Citation

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

@article{he2025the,
      title={The devil is in the details: Enhancing Video Virtual Try-On via Keyframe-Driven Details Injection},
      author={He, Qingdong and Chen, Xueqin and Pan, Yanjie and Tang, Peng and Xu, Pengcheng and Gan, Zhenye and Wang, Chengjie and Hu, Xiaobin and Zhang, Jiangning and Wang, Yabiao},
      journal={arXiv preprint arXiv:2512.20340},
      year={2025}
      }