SiCo: An Interactive Size-Controllable Virtual Try-On Approach for Informed Decision-Making

TL;DR

SiCo tackles online clothing return rates by delivering a web-based, size-controllable VTO that preserves user identity during image-based garment synthesis. The system combines an identity-preserved garment remover, a size-controllable garment mask generator, and a garment generator built on Stable Diffusion with IP-Adapter to produce realistic, size-adjusted visualizations. A user study with 48 participants shows that size-controllable VTO improves users’ ability to visualize fit, understand garment appearance, and increase confidence in sizing decisions, with favorable TLX and SUS profiles. Limitations include sample diversity, potential residual biases, and latency, guiding future work toward broader deployment, enhanced styling options, and faster generation. The work demonstrates the practical viability of integrating size-aware VTO into mainstream shopping experiences to reduce returns and improve informed purchasing.

Abstract

Virtual try-on (VTO) applications aim to replicate the in-store shopping experience and enhance online shopping by enabling users to interact with garments. However, many existing tools adopt a one-size-fits-all approach when visualizing clothing items. This approach limits user interaction with garments, particularly regarding size and fit adjustments, and fails to provide direct insights for size recommendations. As a result, these limitations contribute to high return rates in online shopping. To address this, we introduce SiCo, a new online VTO system that allows users to upload images of themselves and interact with garments by visualizing how different sizes would fit their bodies. Our user study demonstrates that our approach significantly improves users' ability to assess how outfits will appear on their bodies and increases their confidence in selecting clothing sizes that align with their preferences. Based on our evaluation, we believe that SiCo has the potential to reduce return rates and transform the online clothing shopping experience.

Figures (10)

• Figure 1: User Info Logging. The user experience begins with a prompt to upload an image of themselves, guided by specific instructions to ensure accurate processing. This step is essential for the system to generate precise and reliable VTO outcomes. Beneath the image upload section, users are asked to input their true sizes (XXS to XL) for tops and bottoms, defined as the garment sizes that provide a regular fit for their body. This information forms the basis for generating VTO visualizations of clothing in various fits, enabling a personalized and accurate try-on experience. Individual image examples in the interface © Pexels.
• Figure 2: Garment Selection. Our interface offers a curated collection of garments, each accompanied by size selection options tailored to the user's true-size input. The design closely mirrors conventional clothing websites, providing a familiar and intuitive shopping experience. Garment images are sourced from the Dress Code dataset morelli2022dress.
• Figure 3: Users can experiment with their selected garments in the "Try-On Items" section, where each item is accompanied by the user's true size (specified at the start of their session) and the garment size, which can be adjusted at any time. The self-image uploaded by the user at the beginning appears in the "Try-On Room" section under "Before Try-On," serving as the foundation for all subsequent virtual try-on (VTO) visualizations. To try on a garment, users simply select their preferred size and click the "Try It On" button to generate the corresponding visualization (Steps 1 and 2). To facilitate multi-garment styling, each visualization includes a "Continue From Here" button. By clicking this button, users can update the "Before Try-On" image to reflect the current visualization (Step 3), enabling the layering and styling of multiple garments (Step 4). Garment images are sourced from the Dress Code dataset morelli2022dress. Model image © Pexels.
• Figure 4: System Backbone. The backbone comprises an identity-preserved garment remover (Fig. \ref{['fig:backbone_garment_remover']}), a size-controllable garment mask generator (Fig. \ref{['fig:backbone_garment_mask_generator']}), and a garment generator (Fig. \ref{['fig:backbone_garment_generator']}). It processes the user's self-image, true size, selected garment, and size to generate a size-controllable virtual try-on result while preserving the user’s physical identity.
• Figure 5: Effect of User Identity Preservation. The generative model used as the backbone of our system is Stable Diffusion rombach2022high. While this model is capable of producing high-quality images and is generalizable to various use cases, it tends to generate slim and muscular body types that may not accurately reflect users' physiques due to biases in its training data. Without our user identity preservation mechanism, the system often alters user appearance in virtual try-on (VTO) visualizations (Fig. \ref{['fig:without_contour']}). In contrast, our approach prioritizes preserving user identity by maintaining the unique contours and proportions of each user’s body. This mechanism effectively mitigates the inherent biases of Stable Diffusion, ensuring that VTO visualizations remain authentic and representative of the user's true physique (Fig. \ref{['fig:with_contour']}). The garment image is sampled from the Dress Code dataset morelli2022dress. The model image © Pexels