生成式人工智能在面料外观仿真上的研究
Research on the appearance simulation of fabrics by generative artificial intelligence
黄海峤 1李采奕 1张昕莹1
作者信息
- 1. 北京服装学院 服装艺术与工程学院,北京 100029
- 折叠
摘要
数字经济对纺织服装产品的数字孪生仿真有更高的要求,服装数字孪生的品质关键在于纺织面料数字化的质量与效率.本文提出了一种基于机器学习的织物仿真方法.以潜在扩散模型为基础,采用LoRA的微调模型方法,以标签化的织物外观图片集为训练集,训练一个织物外观仿真的模型.与数字服装领域通过扫描面料获得其外观图片的方法相比,该方法速度快、效果好.与成熟的商用图片生成程序生成的图片相比,该模型生成的图片更具有针对性,仿真效果更加逼真.该模型生成的织物外观图片丰富多样,能够根据不同的文本提示词生成不同的织物外观图片,提高了织物外观的设计效率,降低了产品的研发成本,为服装行业的数字化发展和企业的智能制造提供了新的思路和参考.
Abstract
The digital twin simulation of textile and apparel products has higher requirements in the era of digital economy,and the quality of clothing digital twin relies on the quality and efficiency of digitalization of textile fabrics.In this paper,a machine learning-based fabric simulation method is proposed.Built on a latent diffusion model and employing LoRA fine-tuning,this method trains a model for fabric appearance simulation using a labeled dataset of fabric images.In the digital fashion domain,this approach offers faster speed and richer effects compared to methods relying on scanning fabric for appearance images.In contrast to images generated by mature commercial programs,the images produced by this model are more targeted and realistic in appearance simulation.The variety of fabric appearance images generated by this model can realize different fabric appearances based on different textual prompts,enhancing design efficiency and reducing product development costs.This provides new insights and references for the digital development of the clothing industry and intelligent manufacturing for enterprises.
关键词
面料外观仿真/生成式人工智能/潜在扩散模型/机器学习Key words
fabric simulation/generative artificial intelligence/latent diffusion model/machine learning引用本文复制引用
基金项目
北京市服装产业数字化工程技术研究中心科研项目(2020A-17)
出版年
2024
NETL
NSTL
万方数据