Improving Diffusion Models for Authentic Virtual Try-on in the Wild

TL;DR

This paper introduces IDM-VTON, a novel diffusion model for virtual try-on that enhances garment fidelity and authenticity by integrating semantic encoding modules and textual prompts, outperforming previous methods.

Contribution

The paper proposes IDM-VTON, a diffusion-based virtual try-on model with dual semantic encoding modules and textual prompts, improving garment detail preservation and authenticity over prior approaches.

Findings

Outperforms previous diffusion and GAN-based methods in garment detail preservation.

Enhances authenticity of virtual try-on images through semantic encoding and textual prompts.

Effective in real-world scenarios with improved fidelity and authenticity.

Abstract

This paper considers image-based virtual try-on, which renders an image of a person wearing a curated garment, given a pair of images depicting the person and the garment, respectively. Previous works adapt existing exemplar-based inpainting diffusion models for virtual try-on to improve the naturalness of the generated visuals compared to other methods (e.g., GAN-based), but they fail to preserve the identity of the garments. To overcome this limitation, we propose a novel diffusion model that improves garment fidelity and generates authentic virtual try-on images. Our method, coined IDM-VTON, uses two different modules to encode the semantics of garment image; given the base UNet of the diffusion model, 1) the high-level semantics extracted from a visual encoder are fused to the cross-attention layer, and then 2) the low-level features extracted from parallel UNet are fused to the self-attention layer. In addition, we provide detailed textual prompts for both garment and person images to enhance the authenticity of the generated visuals. Finally, we present a customization method using a pair of person-garment images, which significantly improves fidelity and authenticity. Our experimental results show that our method outperforms previous approaches (both diffusion-based and GAN-based) in preserving garment details and generating authentic virtual try-on images, both qualitatively and quantitatively. Furthermore, the proposed customization method demonstrates its effectiveness in a real-world scenario. More visualizations are available in our project page: https://idm-vton.github.io