模型描述 (Model Descriptio)

ELITE是一种基于扩散模型实现图像主体保ID内容迁移的算法。该方法借助基于训练的ecoder提取原图的语义信息，从而实现保ID的文生图和图像内容迁移。和其它方法相比，该算法能通过更快的编码过程实现更高保真度的图像编辑功能。更多详情可参照主页

?Paper |

运行环境 (Operatig eviromet)

Depedecies ad Istallatio

# git cloe the origial repository
git cloe https://github.com/csyxwei/ELITE.git
cd ELITE

# Create a coda eviromet ad activate it
coda create - elite pytho=3.9
coda activate elite

# Istall from requiremets.txt
pip istall -r requiremets.txt

代码范例 (Code example)

参数说明：image为完整输入图像，mask为输入图像经主体分割后的二值图像。toke为占位标记，用以指代目标主体，正常情况下无需修改。prompt为文本输入用以控制图像生成，其中涉及目标主体的词需用toke代替。

from modelscope.pipelies import pipelie
from PIL import Image
import os
import umpy as p

iput = {'image': 'https://vige-video.oss-c-shaghai.aliyucs.com/VideoEhacemet/Dataset/ELITEDataset/1.jpg?OSSAccessKeyId=LTAI5tQUr7s9wUCuorYPo5Qj&Expires=360001692773758&Sigature=mgRwOPaReU1JjAHcC5DEeho904%3D', 'mask': 'https://vige-video.oss-c-shaghai.aliyucs.com/VideoEhacemet/Dataset/ELITEDataset/1_bg.pg?OSSAccessKeyId=LTAI5tQUr7s9wUCuorYPo5Qj&Expires=3600001692773885&Sigature=sTtkRdkc%2BE6EDmtdN%2BBZiRmyN7I%3D', 'toke': 'S', 'prompt': 'a photo of a S'}

iferece = pipelie('my-ELITE-task', model='aojie1997/cv_ELITE_image-cotet-trasfer', model_revisio='v1.0')
output = iferece(iput)
Image.fromarray(output).save('./result.pg')

Citatio

If you fid our work helpful for your research, please cosider citig the followig BibTeX etry.

@article{wei2023elite,
  title={ELITE: Ecodig Visual Cocepts ito Textual Embeddigs for Customized Text-to-Image Geeratio},
  author={Wei, Yuxiag ad Zhag, Yabo ad Ji, Zhilog ad Bai, Jifeg ad Zhag, Lei ad Zuo, Wagmeg},
  joural={arXiv preprit arXiv:2302.13848},
  year={2023}
}

模型描述 (Model Description) ELITE是一种基于扩散模型实现图像主体保ID内容迁移的算法。该方法借助基于训练的encoder提取原图的语义信息，从而实现保ID的文生图和图像内容迁