Behind the Scenes of TryOnDiffusion

Too Long; Didn't Read

The proposed method for virtual try-on utilizes images of a person and a garment worn by another person to synthesize a realistic visualization. The process begins with preprocessing, which includes predicting human parsing maps and 2D pose keypoints. The images are transformed into clothing-agnostic representations, ensuring minimal leakage of garment information. Additionally, the use of diffusion models facilitates an iterative denoising process for generating high-quality outputs conditioned on various inputs.