Size-Variable Virtual Try-On with Physical Clothes Size

TL;DR

This paper introduces size-variable virtual try-on, a framework that adjusts the image size of clothing to match the physical sizes of garments relative to a reference person. It centers on a residual silhouette approach via a Mask Deformation Network (MDN) and a dedicated Texture Fusion step, enabling explicit control over garment size using size vectors. A new dataset with paired images across multiple sizes and a Size Evaluation Metric (SEM) are proposed to quantify size accuracy in the torso and sleeves. Quantitative and qualitative results show improvements in size-consistent rendering (SEM) while maintaining competitive perceptual quality (LPIPS, FID) compared with traditional warping- and IST-based VTO methods. The work enables more realistic, size-aware virtual try-on with potential impact on e-commerce and garment visualization.

Abstract

This paper addresses a new virtual try-on problem of fitting any size of clothes to a reference person in the image domain. While previous image-based virtual try-on methods can produce highly natural try-on images, these methods fit the clothes on the person without considering the relative relationship between the physical sizes of the clothes and the person. Different from these methods, our method achieves size-variable virtual try-on in which the image size of the try-on clothes is changed depending on this relative relationship of the physical sizes. To relieve the difficulty in maintaining the physical size of the closes while synthesizing the high-fidelity image of the whole clothes, our proposed method focuses on the residual between the silhouettes of the clothes in the reference and try-on images. We also develop a size-variable virtual try-on dataset consisting of 1,524 images provided by 26 subjects. Furthermore, we propose an evaluation metric for size-variable virtual-try-on. Quantitative and qualitative experimental results show that our method can achieve size-variable virtual try-on better than general virtual try-on methods.

Figures (11)

• Figure 1: Comparison between previous methods and our method. (i) Previous methods vitonacgpn only generate the try-on image in which the clothes size is the same as the one of the reference person. (ii) A user can change the try-on cloth size in our method.
• Figure 2: Overview of warping-based virtual try-on methods. The try-on mask is estimated from the auxiliary and clothes images. The clothes are warped to fit the mask and integrated with the reference person image to generate the try-on image.
• Figure 3: Two different approaches of mask deformation. (a) Extension of previous methods. The size of the try-on clothes is used for training MDN with auxiliary images in previous methods. (b) Our method. The paired person images in which each person wears different sizes of the same clothes are used in training.
• Figure 4: Sample images of 14 clothes.
• Figure 5: Posture matching for collecting image pairs, in each of which each subject's postures are similar. This matching is done with the Mean Squared Error (MSE) computed between the sets of several body key-points.