DiSAIn-VTON: Diffusion-based Similar Apparel Inpainting Virtual Try-on Network
Published in , 2024
A virtual clothing try-on network using artificial intelligenceis a technology that virtually dresses desired clothing onto a body image. Despite demonstrating high performance, proposed models face challenges in terms of generalization that depend on the training data. The training data used in the network consists of paired images of full-body individuals and clothing items to be worn. However, images of clothing with removed backgrounds, taken in specific poses, can exhibit strong performance during training and validation processes. Yet, obtainable possess diverse poses and include backgrounds, leading to performance degradation. This paper proposes the DiSAIn-VTON model, which applies Clothing Bbox Module and Clothes Geometric Transformation to the Inpainting technique used in the Diffusion model, achieve high generalization performance for clothing encompassing diverse poses and backgrounds. In this paper, experiments are conducted using a newly collected dataset, and through the experiments, the excellence of the model is demonstrated both quantitatively and qualitatively.
Recommended citation: Hyun-woo Jin, Dong-oh Kang, Byung-gook Lee, DiSAIn-VTON: Diffusion-based Similar Apparel Inpainting Virtual Try-on Network, KIISE Transactions on Computing Practices, Vol. 30, No. 3, pp. 149-154, 2024. 3
Download Paper