How We Started Producing Data No One Asked For. At least then…

“Prediction is very difficult, especially about the future”: How We Started Producing Data No One Asked For. At least then…

When it comes to ML, when you start as early as Bones did, it’s hard to learn from mistakes other than your own. Niels Bohr’s famous observation about the difficulty of prediction has never been more relevant than in the rapidly evolving field of artificial intelligence.

Back in 2019, so long before ChatGPT was even a thing (can you even remember that?), we had an idea with our partner for a motion capture dataset with future AI projects in mind. So we did, and it was ready just in time for the recent AI breakthrough.

Anticipating the needs of future RigPlay customers, we were lucky to get it right. Along the way, we gained valuable experience and insights and now that foundation gives us a tailwind as the industry moves forward.

Our most important lessons, apart from a good dose of luck:

-> Utilize the power of professional actor <-

Their skill can make or break the quality of your data. Whether it’s emotional subtleties and variations, knife fight, shooting a bow or just a neutral walk, bringing in professionals is the key due to their range and ability to exaggerate.

-> The energy of directors is the key ingredient <-

A good director can keep up the high level of energy of the team and actors throughout the entire day of shooting, even when you are on the last take of the day. Trust me as an actor myself – recording hundreds of animations per day is not easy on you. Let your directors get the best out of people.

-> Improve your technical excellence one step at a time <-

Allow your technical team to seek improvements constantly. And if they tell you something is complex, and there is no simple and quick answer, most likely they are right – but don’t let perfection get in the way of progress. Move fast, believe in their ideas, iterate!

-> Describe and label everything – not only the movements performed but the costume, conditions, any tiny change you made <-

From the movements performed to the smallest details, every piece of context matters when you are debugging and improving. You never know what will happen to make a difference in the machine learning process. In this context – the more, the merrier.

Along the way, we developed processes and tools that have since become benchmarks in the industry. Today, our datasets are powering some of the most advanced AI systems in the world.

And as we reap the rewards of our early bets and risks taken, our focus is already on what’s next.