XMAS Project — Part 2: Real or Fake? How I got inspired by one slide
October 2020 I got an introduction to a founder called Anna F. Michel. Anna who is Entrepreneur by Heart and someone who identified an industrial issue, which she has faced by her own.
I’m talking about the Design Process in the Textile (especially Fashion) Industry. Some of the biggest Fashion companies are doing about 300.000 designs per year. They spent 2–3% of their GMV for the process from idea to production and they don’t know if it will be a success. yoona.ai tackles this part, taking existing designs, paired with trends and sales data to generate new designs. The goal is to generate designs, which have a higher probability to get bought (based on data) and to digitise the process.
When she told me about the process, I just learned how old the industry is and how much disruption potential. I’ve joined the Advisory Board of yoona.ai and could watched the progress the team has done. During one of our sessions, one slide got my attention.
Which one is fake? Hmmm … guess what? Just one is real. The others are generated by an AI assistant.
It uses a similar technology like https://www.thispersondoesnotexist.com/, https://thisanimedoesnotexist.ai/ or a collection of all of them https://thisxdoesnotexist.com/.
These pages as well as yoona.ai are using GANs (Wikipedia). Read more about it https://medium.com/@danielmanzke/status-quo-getting-into-ai-ml-dnns-21e43a7594e2
The most important thing is data quality. If you take a pie, apple, boat and an airplane, I’m not sure what a GAN would generate?
If you want to generate new designs, you will need a bunch of images. As said yoona.ai pulls trend & sales data for products, gets the images and allows to generate endless new ones based on it.
The team did lately a project for Gerry Weber (Press Release) and I could catch a sneak peak for you. Which ones are Real or Fake?
The team did a tremendous job in the last year. To get there, they had a rocky road. ’cause you have to have the right data and I think this is one of the key assets. Getting the data, filtering, pre-processing and training with it.
The designer will never know what has to be done behind the scenes.
During my XMAS project we will go through the different steps of getting the data, using machine learning to process and hopefully will be able to create something what doesn’t exist.
Stay tuned!
