Image to Construction Pattern
|Coordinator||Stockholms universitet - Institutionen för data- och systemvetenskap|
|Funding from Vinnova||SEK 1 054 975|
|Project duration||November 2017 - December 2018|
Purpose and goal
We provide a demonstrator that shows how to use publicly available images from fashion retailers’ websites and automatically re-engineer physical construction pattern and then map the specific garment to a 3D avatar. We did it by combining existing training data sets with a small native set, and also build a 3D avatar and a show room. The result is promising and shows the potential of this technology to address the fitting-problem in online fashion sales.
Expected results and effects
The goals below are technically oriented and challenging: The identification of fashion relevant key points, was completed. We created a specific data set based on those points. We trained a network that utilize our own data set, in combinations with available fashion data sets, to find our wanted key points. The accuracy of the network is promising, although not at the level where we can integrate it with the other elements. We created a 3D-show room and an avatar.
Planned approach and implementation
The project demanded collaboration between all partners, and in the end they have all contributed. The project was technically demanding, most in terms of whether the AI-component could deliver. It was possible to build technology that almost satisfied our demands. Further, there is also a high demand of researchers with the required skill sets to do these tasks. Here, Stockholm University delivered by recruiting a new Ph D student. The collaboration has also sparked both research and commercialization, which will lead to more results in the future.