基于DreamBooth的傣锦图案人工智能生成模型

Artificial Intelligence Generation Model for Dai Brocade Pattern Based on DreamBooth

为推动传统文化引领下的现代纺织图案设计发展,提出了一种利用人工智能生成民族纺织图案的模型。以傣族织锦图案为例,对傣族织锦实物图案进行矢量化处理,为每张图片编写对应文本标签,并以此为训练集;选取适用于傣锦的预训练模型,使用DreamBooth方法微调现有的文本-图像模型。对模型训练效果进行分析,得出V1模型是一个具有良好拟合度和图像生成效果的文本-图像生成模型。

In order to promote the development of modern textile pattern design under the guidance of traditional culture,a model of introducing ethnic textile patterns into the field of artificial intelligence creation was proposed.Taking the Dai brocade pattern as an example,the physical images of Dai brocade were vectorized,and the corresponding text labels were written for each image,and the training set was used.The research selected a pre-trained model suitable for Dai brocade,and used the DreamBooth method to fine-tune the existing text image models.By analyzing the training effect of the model,it is concluded that V1 model is a text image generation model with good fitting degree and image generation effect.