No two humans are alike. We all look, and move, a bit differently. In fact, some of the most compelling work I've ever had the chance to encounter was on the representation of race within research populations in science, presented by Dr. Gillian Mazzeo, one of our team members and a fantastic scientist in her own right. In those conversations, it was really brought to light how important it is to have data that represents the actual diversity that exists within a given population, and for good reason. If all technology is trained purely on data from one demographic, it’s going to be a poor representation of the diversity that exists in the real world, which is really important to us here at Theia.

From the start, we have trained our deep learning algorithms to be as representative as possible of the full diversity of human body types. Because markerless technology works by identifying and tracking points on the body in video, the data used to train the algorithm must adequately represent the humans the algorithm will be tracking. To do this, we worked to source very large, diverse datasets in order to train the algorithm to work across body types and conditions. The result is a deep learning model that is trained across body type, race, gender, activity, and appearance. That training data is then coupled with data from the labs of our partners in the clinical and research world to produce an algorithm that is well adapted for any human movement scenario.

The result is that Theia3D can work accurately across body types and demographics. This is something that we think is really important, and we hope that you do too. If you’d like to discuss how our system could work for your work, please reach out here.