ComfyUI_CSD_MT

★ 19

化妆迁移无监督学习风格内容解耦ComfyUI节点

在ComfyUI中实现无监督化妆迁移，采用内容-风格解耦方法，无需生成伪真值即可改变妆容风格并保留原图结构。

💡 在ComfyUI内对人像进行无监督化妆风格迁移

🍴 1 Forks💻 Python🔄 2025-02-06

🔗 GitHub 原文

📦

网盘下载

复制链接后前往夸克网盘下载

https://pan.quark.cn/s/c1eafc754fbb

📦 requirements.txt

torch
torchvision
numpy
opencv-python
imageio
tensorboardX
cmake
#build
for
dlib
boost
#build
for
dlib
dlib

📄 README

ComfyUI_CSD_MT

CSD_MT is a method about ‘Content-Style Decoupling for Unsupervised Makeup Transfer without Generating Pseudo Ground Truth’, you can use it in comfyUI.

Tips

This method need little space and VRAM(or CPU),all the models is about 200M(4 weights),and best quality size is 256*256

这个方法所用4个模型加起来不到200M,所需显存或CPU可以忽略不记,当然,图片输出最佳质量是256*256.

1. Installation

In the ./ComfyUI/custom_node directory, run the following:

git clone https://github.com/smthemex/ComfyUI_CSD_MT.git

2. Requirements

pip install -r requirements.txt

When install dlib get error,use wheel to install it,download wheel from this address

如果按照dlib库失败，从此链接下载对应你python版本的轮子，用以下命令安装：

pip install dlib-19.24.1-cp311-cp311-win_amd64.whl

3.Model

3.1 download 3 checkpoints (79999_iter.pth,CSD_MT.pth,lms.dat) from google or baidu password:1d3e.从百度云或者Google下载三个模型

3.2 download ‘resnet18-5c106cde.pth’ from here 从链接下载resnet18-5c106cde.pth模型，不要改名字.

--  ComfyUI/models/CSDMT
    |-- 79999_iter.pth
    |-- CSD_MT.pth
    |--resnet18-5c106cde.pth #没有也会自动下载
--  ComfyUI/custom_node/ComfyUI_CSD_MT/quick_start/faceutils/dlibutils
    |--lms.dat

4.Example

5.Citation

@inproceedings{sun2024content,
  title={Content-Style Decoupling for Unsupervised Makeup Transfer without Generating Pseudo Ground Truth},
  author={Sun, Zhaoyang and Xiong, Shengwu and Chen, Yaxiong and Rong, Yi}
  journal={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
  year={2024}
}