Omnious Unveils Apparel AI Technology at World’s Leading ECCV Conference

Virtual Images Realistically Wearing Models Created by AI
Research Conducted with KAIST

Artificial intelligence (AI) company Omnious announced on the 30th that it will present a research paper on virtual clothing try-on AI technology at the International Conference on Computer Vision, the European Conference on Computer Vision (ECCV), held in Milan, Italy.

Held every two years, ECCV is the world's most prestigious conference where the latest research on AI technologies specialized in image and video analysis, along with the field of computer vision, is presented. Global big tech companies such as Google and Meta, as well as leading scholars in AI, participate in large numbers.

Omnious conducted the research jointly with the laboratory of Professor Shin Jin-woo, technical advisor and endowed professor at KAIST AI Graduate School, and passed the rigorous review process of ECCV, which has an acceptance rate of only about 27.9% this year, thereby gaining recognition for its AI technology capabilities.

The paper adopted this time by Omnious is titled “Improving Diffusion Models for Natural Virtual Try-On in Real Environments,” proposing a method to reproduce clothing as if actually worn through AI without the model needing to physically wear the clothes.

To this end, Omnious applied a dual analysis approach based on ▲ a module that comprehends overall clothing characteristics such as style and color, and ▲ a module that identifies detailed features such as patterns, textures, wrinkles, and decorations, producing research results that offer more natural image synthesis compared to existing image generation models.

Omnious plans to introduce an AI service to major domestic shopping commerce platforms within the year that generates everything from clothing to models and backgrounds using only product images and text descriptions by leveraging AI technology.

Choi Hyung-won, head of Omnious AI Research Center, said, “The distinction lies not only in reflecting the shape but also vividly expressing shadows and wrinkles according to the model’s body shape and pose,” adding, “We will continue to advance the technology to help improve the online shopping environment for brand clients through realistic image generation.”