[ECCV 2024] Time-Efficient and Identity-Consistent Virtual Try-On Using A Variant of Altered Diffusion Models 

We have developed a novel framework for time-efficient and identity-consistent virtual try-on, leveraging an altered variant of diffusion models to advance e-commerce and metaverse applications. This work, “Time-Efficient and Identity-Consistent Virtual Try-On Using A Variant of Altered Diffusion Models,” was accepted at ECCV 2024, a premier A conference in computer vision with an acceptance rate of 27.9%.

Virtual try-on is a critical challenge in digital fashion, requiring not only realistic synthesis of clothing textures but also strict preservation of personal identity features such as facial traits, tattoos, or accessories. Existing approaches often struggle with identity omission, uncontrollable artifacts, and slow synthesis speeds, which limit their practicality. Our study addresses these gaps by introducing a diffusion-based pipeline designed to simultaneously preserve garment textures and user identity while achieving unprecedented speed.

The proposed framework consists of two key components: a warping module, which aligns clothing with individual features, and a try-on module, which refines attire generation and reconstructs missing details. This process is enhanced with a mask-aware post-processing technique that safeguards the integrity of identity-specific features. Experimental validation demonstrates that our method delivers high-quality results with remarkable efficiency—nearly 20 times faster during inference compared to state-of-the-art baselines—while maintaining superior fidelity in qualitative evaluations. Quantitative assessments on the VITON-HD and Dresscode datasets further confirm that our approach achieves performance comparable to the strongest existing methods.

The potential applications of this work span shopping kiosks, online fashion platforms, and immersive metaverse experiences, where speed, realism, and user identity retention are all essential. By bridging quality and efficiency, this research marks a major step toward bringing high-quality virtual try-on systems into real-world deployment.

This research is the very first step to bring high-quality VITON to the real world, enhancing the e-commerce fashion experience.