Virtual Try-ON Using Generative Adversarial Networks (GAN)

Background

Recent years have seen a significant rise in online fashion shopping. But despite the convenience online retail provides, consumers are often concerned about how items may look or fit in person. This has led to a culture of returns that has a significant impact on carbon emissions. Therefore, allowing consumers to try on clothes virtually will not only enhance their shopping experience, but also help retailers to reduce waste and their impact on the planet.

Image-based virtual try-on systems have attracted increasing research attention in recent years, yet it remains a challenging field. An ideal solution should not only seamlessly fit clothes to the users body, but also preserve the look of the clothes in the generated image including the texture, logo, and any embroidery or details etc.

Objective of this Experiment

The primary objective of this experiment is to develop a simple and effective GAN architecture and training strategy that generates photo-realistic try-on results, while preserving both the character of clothes and details of human identity (posture, body parts, other items of clothing, etc.).

For the training purposes, we used a VITON dataset created specifically for this purpose. It consisted of 16,253 pairs of images, each containing a person and a top-half clothing item.

Business Use Cases and Applications

There are multiple direct and indirect applications of this experiment. Some of them include:

• Previewing makeup and hair colours: An online virtual try-on feature is extremely helpful for testing out different kinds of hair colours and cosmetic products. These tools can examine complexion and produce a specific and perfect signature colour for an individual’s lips, face, eyes and hair. And, in turn, help the customer to select the most suitable products.
• Choosing the best pair of glasses: Similarly, virtual try-on features are often used by people who struggle to find the perfect pair of glasses for their facial features.
• Creating a virtual dressing room: The virtual dressing room lets the users try out different outfits, such as skirts, shorts, tops and accessories, and build the best ensemble for every occasion or season.