Building Video-Understanding AI: How We Taught Computers to Watch Sports
by Varun Kodathala and Prashanth Pollishetty | at Minnebar 19
In this session, we'll walk through our technical approach to creating a video understanding system specifically for sports analysis. We'll share insights from our development of a lightweight yet effective framework that helps computers understand and analyze sports videos.
Key topics we'll cover:- How we tackled the challenge of quickly identifying the most important moments in sports videos
- Our approach to balancing detailed visuals with motion analysis
- How we extract key frames that represent critical positions in sports movements
- Methods we developed to preserve the dynamic movement between these key frames
- Our results: 60.8% improvement in cross-modal retrieval accuracy and 9.7% higher correlation in semantic textual similarity compared to standard benchmarks
- How our specialized sports language model outperformed models 5 times its size in accuracy and relevance metrics
- Our resourceful development process using only two GPUs (thanks to Microsoft) to build a high-performing system
- Practical challenges we faced with amateur sports videos, including varying camera angles and lighting conditions
This session is for those interested in computer vision, sports technology, and practical applications of AI in resource-constrained environments. We'll share both our technical achievements and the real-world limitations we encountered along the way.
Varun Kodathala
I am Sai Varun Kodathala, the CTO at SportsVision (Techstars' 24). My background in engineering, highlighted by two University President Gold Medals, has grounded my approach to technological solutions. I specialize in image processing and computational optimization, focusing on practical applications in sports analytics. My work includes developing algorithms that enhance our understanding of athlete biomechanics and performance. I have also filed provisional patents to secure our innovative methods. Passionate about bridging the gap between academic theories and real-world applications, I am committed to advancing sports technology through thoughtful and effective solutions.
Links:
Prashanth Pollishetty
Prashanth Pollishetty is the CEO and Founder of Vizworld and SportsVision, where he skillfully merges strategic foresight with profound technical expertise, particularly in the sports analytics sector. Previously, he served as the Senior Director of Technology at United Health Group, showcasing his ability to manage complex data systems effectively. His tenure at Wells Fargo as Vice President of Strategy further highlighted his innovative approaches to technology integration, driving significant business advancements. Prashanth's visionary leadership was instrumental in securing a groundbreaking project with the USOPC, demonstrating his capacity to steer his team toward innovative solutions in the sports technology landscape. His foresight in anticipating market trends continues to guide his company toward pioneering advancements.
Are you interested in this session?
This will add your name to the list of interested participants. It will help us gauge interest for scheduling purposes.
Interested Participants
Similar Sessions
Does this session sound interesting? You may also like these: