One of the root causes for the skewed gender participation in skateboarding is the misconception that ‘girls are not good enough’ for extreme sports. Especially in disciplines such as skateboarding where judging criteria such as ‘style’ often leave critical points up to interpretation, in the minds of peers, laypeople and judges, as well as in competition and across the over 5000 skate parks globally. By using regular camera footage and building technology that helps identify complex tricks for the first time, Project Skate helps bring objectivity to a highly subjective field and it connects the technologies of Tensorflow and Blazepose in a brand new way, uniquely suited for skateboarding and its users.

We cannot change what we can’t measure. Skateboarding is a highly biased sport, that has a huge growth rate but overwhelmingly male, and a culture that works to exclude women and girls. This is why Project Skate aims to provide skateboarders of all levels with a tool that allows them to prove their ability objectively - regardless of gender, nationality, age or other factors that might make them susceptible to biases. Blazepose and Tensorflow make skateboarding data visible for the very first time, encouraging participation and learning. To test our tool and teach it trick detection, Olympic medalist Sky Brown was the first professional skateboarder to test out the prototype and to prove her awesomeness with the data she had never seen before.

Project Skate can become a role model on ethical, productive and truly beneficial use of technology that does what humans can’t - instead of replacing their subjectivity. The launch film was presented at I/O and viewed 31.5 million times by 17.7 million people; 97% of all comments were positive. There is an appetite to defeat biases with facts and equalize the playing field - and skateboarding is a hugely complex, freestyling sport that can pave the way for other versions of this tool. Hence the potential to scale into other, mostly less complex sports, as the logical next step.

The tool tracks the coordinates of 33 positions on a skater's body, 30 times a second. With three cameras - and assuming a length of 5 seconds - a single trick contains over 14,000 data points. Powered by Google’s open-source machine learning tools (Tensorflow and Blazepose), Project Skate uses the power of data to measure athletes’ performances objectively. A simple setup of three cameras generates all data necessary to define an athlete’s position in 3D space at any given moment. Trained on hundreds of hours of skating footage, Project Skate then measures their speed, height, rotation, grabs and movement of the skateboard. These combine to give a definitive view of the trick performed and its attributes. The lovingly crafted UI provides real-time feedback inspired by vintage skateboarding video games. The working MVP has been tested by Sky Brown and succeeded in recognizing even tricks & moves it had not been taught.