Akzo nobel

Even for the most experienced paint specialist, pinpointing which paint defect you’re dealing with can be time consuming. That’s why we conceptualized, designed and developed a mobile web app that identifies defects and provides step-by-step guides for fixes and prevention. The app uses machine learning to match the paint defect in uploaded photos to an extensive database. To create this fully automated solution, we developed a first-party data set to train the application with digitally simulated paint defects. With every user-uploaded photo, our paint-repair tool becomes smarter and more accurate—saving specialists time and increasing efficiency.

A problem came to us by Akzo: How can we use a computer/app to identify any paint defect? They have been spending a lot of money on sending people to identify various defects or there customers are having to spend extra to get it identified and then continue on the corse of action to fix it. The idea was to be able to empower the customer with a tool in their pocket to identify the issue and supply a course of action to fix it. We needed to build a CNN Knowledge Model for all the paint defects. So the question remained how were we going to collect all the data necessary to build it. We needed many pictures in various environmental conditions and forms just to detect one defect. Realizing this was something that would be quite difficult we needed to take another approach. So we requested some physical samples of the defects from the client, brought them to our VFX team and built an engine changing variables to match all the different conditions we would expect. This gave us the ability to generate thousands of simulated pictures for each paint defect to train the model, and it worked!