We are a group of tech enthusiasts and researchers from Berlin - a city that turns into Siberia in winter and Tuscany in summer.
We experienced first hand that the training of A.I. algorithms not only relies on the availability of relevant data. It also gets stuck in development cycles for years when the outcome does not meet the expectations.
A few years ago we started a computer vision project with millions of real-life images at our disposal. However, as it turned out our holy grail was not so holy after all. We were not able to use most of the available data due to bad quality, bias problems or privacy concerns. We faced the prospect of months of additional data collection and annotation.
At one point we realized that we could use our previous experience in photogrammetry and ultra-realistic 3D live renderings to create exactly the data we need.
The result were fully synthetic datasets that performed exceptionally well. In the process, we succeeded in bridging the notorious "domain gap". As a result, Promethion was born.
Developers should not waste time on data sourcing and algorithm optimization. It makes much more sense (and fun!) to focus on making a UAV fly and a robot fulfill the task at hand.