Sports 2i Builds a Highly Accurate Computer Vision Dataset from Scratch with Superb AI
Sports 2i wanted to expand its research and operations into artificial intelligence to analyze player movements and identify areas of improvement. They needed to build an initial training dataset for their model but didn’t have a data management process in place.
Superb AI’s personalized data services through an initial three-month term.
~99% accuracy rate of ground truth dataset
100% POC satisfaction reported
The growing popularity and competitiveness of professional sports leagues have led to an increased emphasis on data analysis for optimizing team performance. Coaches and staff must rely heavily on game data to find an edge, develop tactical strategies, and identify team strengths and weaknesses. This data can also help teams with less funding offset the cost of first-tier players by more effectively scouting and developing second-tier players.
Sports 2i, a sports big data specialist, chose Superb AI to help them integrate artificial intelligence into their research and business operations. Before working with Superb AI, Sports 2i had yet to develop a training data management process and needed to develop one from scratch. Building their first training datasets and establishing an infrastructure for model development thus became the top priority.
This was, of course, easier said than done. Creating a dataset to predict players’ patterns involved tracking both movements and ball possession and would require a lot of data and annotation types. This includes classifying play types, such as pick-and-roll, post-up, isolation, tracking acceleration and deceleration, and identifying optimal shooting opportunities and positions. This information, when processed and fed to the model, assists coaches and analysts in optimizing player performance and developing tailored training programs.
"We requested the project at the end of the year when everyone in the industry was busy, so despite our efforts to reach out to many companies, including Superb AI, with highly detailed specifications on raw data and labeling guidelines, many of them declined the project because of lack of resources or asked for too much time for feasibility checks.
Also, even though we know they are not technicians, it was important for us that we communicate with sales representatives and project managers who have high levels of understanding of AI technologies and the industry - and it was really difficult to find a company that satisfies these conditions.
That's when Superb AI expressed their willingness to proceed with the project and diligently conducted pre-project planning and a PoC. What impressed us the most was that, when we suggested requirements that Superb AI's platform or staff couldn't accommodate, they didn't just say no but instead suggested a satisfying alternative within one business day."
Solving these challenges is where Superb AI’s project managers came in - helping to set up the direction, workflow, and strategy. Over three months, Superb AI proceeded to help label and audit Scale 2i’s raw data and provide strategic advice and personalized support, even when challenges arose. Hyowon Park, Converged Business Team Lead for Sports 2i, says of his experience working with the Superb AI team: "They were always very willing to accommodate our requirements as much as they could, making us feel that the team is willing to go the extra mile to make our project successful."
The project ultimately proved a big success for Sports 2i, netting them a 99% accuracy rate across all object types and classes within their ground truth dataset. When subtly asked if his team was planning another project, he confirmed that they would definitely use Superb Services again. He expressed appreciation for Superb AI's emphasis on the quality and detail of the dataset rather than solely on the speed of completion.