The machine learning tool fits the garment to the shape and pose of a person’s image
With the coronavirus pandemic keeping many physical retail stores shuttered and making photo shoots of all kinds more difficult, Adobe is experimenting with new ways for shoppers or models to virtually try on clothes. The software company outlined a new potential feature this week that uses artificial intelligence to transform a piece of clothing to fit the shape and pose of a person in an image while preserving the visual details of the garment, according to a research paper Adobe compiled with Stanford University and IIT Hyderabad.
The researchers claim their method can outperform other similar tech, which often suffers from unsightly distortions, by first warping the apparel to the body and then applying the texture of the warped item to the target model in a way that more seamlessly integrates with the image in question. The system is trained on around 19,000 images of female models and product listings. “While we don’t expect physical photo shoots to go away, there will be instances where the time and cost savings can be compelling—especially when the right technology is available,” Adobe senior data and analytics evangelist Eric Matisoff said in a blog post. “It aligns with Adobe’s long history of working with graphics and AI that has been trained to understand the nuances in composition, textures and the like.”
As image recognition and generation AI has improved in recent years, a number of similar tools based on similar technology have already hit the market, including startups like Vue.ai and Zeekit and L’Oréal’s virtual makeup counter ModiFace. Adobe researchers claim their tool could improve on many of these existing offerings. “We show significant qualitative and quantitative improvement over the current state-of-the-art method for image-based virtual try-on,” they said in the paper. “Successful virtual try-on experience depends upon synthesizing images free from artifacts arising from improper positioning or shaping of the try-on garment, and inefficient composition resulting in blurry or bleeding garment textures in the final try-on output.” Adobe imagines the feature eventually being adaptable to the needs of a specific client brand, whether they are looking to power a virtual fitting room feature or interchange models for product listings. The company is also promoting it as a possible way to increase the diversity of models featured on ecommerce sites.