首页|基于Diffusers框架的电商产品图制作方法研究

基于Diffusers框架的电商产品图制作方法研究

扫码查看
基于Diffusers框架与Control Net网络,文章开发并实现了服装产品图的制作流程,将Segment Anything与Grounding DINO结合,实现了自动分割衣服蒙版.将Semantic Segmentation与ViTmatte结合,实现了对皮肤的精细化分割.文章使用大模型的低秩适应实现了对图像风格的微调,利用Ultralytics框架,结合Diffusers提出了一种图像修复方法.将生成图与原图对比,实验结果表明:将Diffusers与Control Net网络结合,能够实现模特替换与背景替换,该方法可以为用户提供衣服上身的参考,降低商家的产品图制作成本.
Research on e-commerce product diagram creation method based on Diffusers framework
Based on the Diffusers framework and Control Net,the article develops and implements the production of clothing product images.The author combines Segment Anything with grounding DINO to achieve automatic segmentation of clothing masks.Semantic segmentation is combined with ViTmate to achieve fine segmentation of skin.Low rank adaptation of large models is used to achieve fine-tuning of image styles,and proposed an image restoration method using the Ultralytics combined with diffusers.At the end of this article,the author compares the generated image with the original image.The results show that by combining Diffusers with the Control Net,it can realize model replacement and background replacement.The method provides reference for users on upper-body clothes and reduces costs of the product diagram creation.

e-commerce product diagrammodel replacementDiffusersControl Net

伍海洋、马川越、刘登辉

展开 >

西京学院 电子信息学院,陕西 西安 710123

电商产品图 模特替换 Diffusers Control Net

2024

无线互联科技
江苏省科学技术情报研究所

无线互联科技

影响因子:0.263
ISSN:1672-6944
年,卷(期):2024.21(2)
  • 8