The Pitt Cloud Innovation Center, powered by Amazon Web Services (AWS), collaborated with the Pitt Diving team to improve operational efficiency and create automated highlight reels. Before this solution, coaches dedicated hours to reviewing practice footage, meticulously searching for individual dives.
Coach Katie Kasprzak, Pitt Diving’s head coach since 2017 and an advocate for athlete development, knew technology was the answer, and came to the Pitt Cloud Innovation Center (CIC) for help.
"I was spending approximately 10 hours per week evaluating practice; manually tracking our athletes performance development and sifting through footage to provide feedback. It was taking up a significant amount of time and I knew there had to be a better way.”
-Katie Kasprzak, Head Coach, Pitt Diving
The Automatic Highlight Reel Generator is an innovative technical solution developed by the CIC that directly addresses this challenge. It's designed to automate the most tedious parts of video analysis for sports teams, starting with Pitt Diving.
The Challenge: Time-Consuming Video Review
Coach Kasprzak spent hours sifting through practice footage and reviewing dives. This meant sometimes searching for specific dives, other times reviewing particular divers. This process typically consumed roughly ten hours per week. She recognized the need for a more efficient way to leverage their practice videos, seeking a solution that could quickly and accurately pinpoint individual dives, transforming video analysis into a tool for efficiency rather than a drain on resources.
The Solution: Automated Video Processing
To address Coach Kasprzak’s challenge, Pitt Cloud Innovation Center interns Roman Koshovnyk and Rowan Morse spearheaded the development of a system that automates the entire process.
Here's how it works:
- Automated Clipping: The system intelligently processes long practice footage.
- Individual Dive Identification: It automatically detects and isolates individual dives.
- Dramatic Reduction in Review Time: Coaches no longer need to manually search for specific moments.
- Focus on Coaching: By automating the mundane, the solution allows coaches to dedicate more time to instruction, strategy, and athlete well-being.
"During our initial practice observation, we saw the team use a delayed live TV for feedback, but the valuable footage wasn’t being saved for later review. As a former athlete myself, I know that for athletes and coaches, the ability to review repetitions is critical. Saved footage allows for detailed, frame-by-frame analysis and comparing progress over time."
-Rowan Morse, Student Intern
With that in mind, the students built a solution that could detect the start and end of each dive and automatically clip the footage into individual repetitions.
Developing a reliable system under real-world constraints like single-angle footage, background footage of coaches and other divers, plus water reflections was a challenge. However, after trying standard computer vision models, the team decided to pivot to a zero-shot vision-language model (VLM) to enable flexible prompting and strong performance with minimal supervision.
The solution is able to accurately find 97% of the dives in the 2.5 hour practice session video footage.
This solution significantly streamlines the workflow, empowering coaches with readily accessible, precisely clipped video segments. “It transforms hours of manual review into just minutes”, says student intern Roman Koshovnyk, "directly contributing to more effective coaching and athlete improvement."
The Industry: Sports Analytics
While the initial success of the Automatic Highlight Reel Generator is with the Pitt Diving team, the underlying principles and technology are highly adaptable to a wide range of other sports. The core idea — automatically identifying and clipping specific events from continuous video footage — has broad applications in athletics and beyond.
This automated approach frees up valuable time, allowing coaches to concentrate on athlete development and strategy — including personalized athlete feedback — ultimately enhancing performance at all levels.
“This solution has completely transformed how we analyze practice, turning it from a massive drain on my time into a powerful tool that provides valuable quantifiable data that tracks the athletes progress in a meaningful and user-friendly manner."
- Katie Kasprzak, Head Coach, Pitt Diving
The Application: Applicability to Other Industries
The open-source framework can be adopted by any team, group, or organization that is interested in identifying specific events from continuous video footage.
Supporting Artifacts
Want to dive deeper into the technical details of this innovative solution? Explore the code and project specifics on GitHub.
