Article: Deep learning based automated generation method for fashion design Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.26 pp.51 - 67 Abstract: Intending to the issue of low fidelity of images generated by existing fashion design methods, this paper firstly segments the original garment images semantically based on VGG and Unet, encodes the target garment part images into different part features, and obtains the appearance flow field of the target garment parts. Then the target garment parts are deformed according to the appearance flow field of the garment parts, and the features of each garment part are fused. Finally, the garment images are generated in light of generative adversarial network (GAN) and diffusion model. The degrees of freedom are restricted by human posture control module and the region degree discriminator is used to enhance the local fine-grainedness of the garment images. The experimental results show that the structural similarity index (SSIM) of the proposed method is 0.895, and the generated results are clearer and more realistic. Inderscience Publishers - linking academia, business and industry through research

Title: Deep learning based automated generation method for fashion design

Authors: Chen Liang

Addresses: College of Art and Design, Zhengzhou University of Economics and Business, Zhengzhou 451191, China

Abstract: Intending to the issue of low fidelity of images generated by existing fashion design methods, this paper firstly segments the original garment images semantically based on VGG and Unet, encodes the target garment part images into different part features, and obtains the appearance flow field of the target garment parts. Then the target garment parts are deformed according to the appearance flow field of the garment parts, and the features of each garment part are fused. Finally, the garment images are generated in light of generative adversarial network (GAN) and diffusion model. The degrees of freedom are restricted by human posture control module and the region degree discriminator is used to enhance the local fine-grainedness of the garment images. The experimental results show that the structural similarity index (SSIM) of the proposed method is 0.895, and the generated results are clearer and more realistic.

Keywords: fashion design; automated generation; deep learning; generative adversarial network; GAN; diffusion model.

DOI: 10.1504/IJICT.2025.147469

International Journal of Information and Communication Technology, 2025 Vol.26 No.26, pp.51 - 67

Received: 14 May 2025
Accepted: 25 May 2025
Published online: 16 Jul 2025 *

Title: Deep learning based automated generation method for fashion design

Keep up-to-date