Virtual Try-On System

Abstract

Online shopping often faces challenges such as size mismatch, poor fit visualization, and high product return rates. To overcome these issues, an AI-powered Virtual Try-On System is designed to improve the online apparel shopping experience. It is a computer vision-based solution that allows users to virtually try clothes before purchasing. The system accepts two inputs: an image of the user and an image of the desired clothing item. It then generates a realistic output image showing the user wearing the selected garment. This is achieved through a combination of image inpainting, human body segmentation, and reference guided image synthesis techniques. Human body segmentation is used to accurately identify and preserve key body parts such as the face, arms, and legs, ensuring that only the clothing region is altered. Image inpainting is employed to seamlessly blend the new garment onto the user's body, maintaining visual coherence. Reference-based conditioning ensures the selected clothing item is appropriately aligned and adjusted to fit the user’s posture and body structure. By combining these techniques, the system creates high-quality virtual try-on results that are visually realistic and anatomically accurate. This project aims to reduce the limitations of physical try-on, improve user engagement, and minimize product returns in e-commerce platforms. It demonstrates the practical application of AI in fashion technology, offering a smarter, more interactive, and personalized online shopping experience.