CAFANet: Cascade Appearance Flow Assemble Network for Virtual Try-On
Liang-Ying Ke, Jocelyn Liao, Chih‐Hsien Hsia · 2024
With the rapid development of deep learning (DL), computer vision (CV), and the internet of things (IoT), virtual try-on (VTON) technology for online clothing shopping has also achieved breakthrough results and is beginning to be popularized in the market. However, VTON technology still makes it difficult to warp complex garments, and texture distortion is common in deformed garments. To address the above problems, this study proposes a cascade appearance flow assembly network (CAFANet) based on global-local appearance flow estimation for VTON. The model effectively avoids the problem of garment texture distortion caused by aligning garment shapes and garment textures by cascade global-local alignment modules (CGLAM). According to the experimental results, the model architecture proposed in this study has good clothing synthesis ability on the VITON database and reaches a FID value of 25.34 on the test dataset.