一种草图生成对抗网络、渲染生成对抗网络及其服饰设计方法
By combining unsupervised sketch generative adversarial networks and rendering generative adversarial networks, this study solves the problems of declining quality in complex texture generation and dependence on supervised information in existing fashion image generation models. It achieves high-quality, personalized fashion item image generation, improving design efficiency and interactivity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH SHENZHEN GRADUATE SCHOOL
- Filing Date
- 2021-12-15
- Publication Date
- 2026-07-17
AI Technical Summary
Existing fashion image generation models suffer from quality degradation when generating images of fashion items with complex textures, require a large amount of supervised information, lack controllability and interactivity in fashion design, and are unable to meet the personalized needs of designers.
An unsupervised sketch generative adversarial network is used to generate sketch images, and combined with a rendering generative adversarial network, realistic fashion item images are generated by utilizing latent space and texture information through a dual discrimination mechanism of sketch discriminator and rendering discriminator. An end-to-end and divide-and-conquer training strategy is designed to optimize the network.
It enables controllability over the generation type and texture during the generation process, improves the quality of generated images and the personalization of fashion designs, reduces the learning cost and time for designers, and enhances the interactivity of user needs.
Smart Images
Figure CN116266251B_ABST