GitHub - levihsu/OOTDiffusion: OOTDiffusion: 基于服装融合的潜在扩散的官方实现,用于可控虚拟试穿

内容

OOTDiffusion

ootdiffusion

这个存储库是OOTDiffusion的官方实现

尝试我们的OOTDiffusion

🤩 如果您觉得有趣,请给我一个星星!

OOTDiffusion: 基于服装融合的潜在扩散,用于可控虚拟试穿
Yuhao Xu, Tao Gu, Weifeng Chen, 和 Chengcai Chen
Xiao-i Research

我们论文的早期版本现已发布!arXiv

🥳🥳 我们基于VITON-HD(半身)和Dress Code(全身)训练的模型检查点已发布!

demo workflow

安装

安装

  1. 克隆存储库

git 克隆 https://github.com/levihsu/OOTDiffusion

  1. 创建一个 conda 环境并安装所需的软件包

conda create -n ootd python==3.10 conda activate ootd pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 numpy==1.24.4 scipy==1.10.1 scikit-image==0.21.0 opencv-python==4.7.0.72 pillow==9.4.0 diffusers==0.24.0 transformers==4.36.2 accelerate==0.26.1 matplotlib==3.7.4 tqdm==4.64.1 gradio==4.16.0 config==0.5.1 einops==0.7.0 ninja==1.10.2

推断

推断

  1. 半身模特

cd OOTDiffusion/run python run_ootd.py --model_path <model-image-path> --cloth_path <cloth-image-path> --scale 2.0 --sample 4

  1. 全身模型

服装类别必须配对:0 = 上半身;1 = 下半身;2 = 连衣裙

cd OOTDiffusion/run python run_ootd.py --model_path <model-image-path> --cloth_path <cloth-image-path> --model_type dc --category 2 --scale 2.0 --sample 4

引用

引用

@misc{xu2024ootdiffusion,
      title={OOTDiffusion: Outfitting Fusion based Latent Diffusion for Controllable Virtual Try-on}, 
      author={Yuhao Xu and Tao Gu and Weifeng Chen and Chengcai Chen},
      year={2024},
      eprint={2403.01779},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

待办事项清单

待办事项清单

  • 论文
  • Gradio 演示
  • 推理代码
  • 模型权重
  • 训练代码
总结
OOTDiffusion is an implementation by Xiao-i Research for controllable virtual try-on. The paper 'OOTDiffusion: Outfitting Fusion based Latent Diffusion for Controllable Virtual Try-on' by Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen is available on arXiv. Model checkpoints trained on VITON-HD and Dress Code have been released. The repository provides installation instructions and inference steps for both half-body and full-body models. The code and models have been tested on Linux. The repository also includes a TODO list for future updates. The implementation uses various packages and checkpoints for preprocessing. A citation for the paper is provided for reference. Additionally, a Gradio demo, model weights, and training code are planned for future updates.