Inspiration
Cricket is a game of timing, technique, and decision-making. Every shot tells a story, but identifying and analyzing shots across long videos can take considerable time. I wanted to explore how artificial intelligence could make cricket video analysis smarter and more efficient. That idea inspired me to build CricketIQ.
What it does
CricketIQ is an AI-powered cricket shot recognition project that uses a fine-tuned VideoMAE model to classify cricket shots from video clips.
The model is designed to recognize ten shot categories:
- Cover
- Defense
- Flick
- Hook
- Late Cut
- Lofted
- Pull
- Square Cut
- Straight
- Sweep
The goal is to make shot identification easier and explore how video understanding can support cricket analysis.
How we built it
I built the project using Python and PyTorch for the machine-learning workflow, with a fine-tuned VideoMAE model for video-based classification.
I developed a FastAPI backend to handle prediction requests and a web frontend using HTML, CSS, and JavaScript. The project brings together video understanding, model inference, API development, and web technologies.
Challenges we ran into
One major challenge was working with video rather than individual images. Recognizing a cricket shot requires learning visual patterns across frames, not just identifying a player or object in a single image.
Deployment was another challenge. Video models require significant memory, and hosting the trained model within limited cloud resources proved difficult. This showed me that building a model is only one part of creating a usable AI application.
Accomplishments that we're proud of
I'm proud to have developed a project that applies a video-understanding model to a specific sports problem. Bringing together a trained deep-learning model, a backend API, and a web interface gave me experience with multiple parts of the AI application development process.
I'm also proud that CricketIQ provides a foundation that can be extended into a broader cricket video analysis tool.
What we learned
This project strengthened my understanding of video transformers, transfer learning, PyTorch, API integration, and frontend-backend communication. It also taught me to consider practical deployment constraints alongside model development.
Most importantly, I learned to approach AI as a complete product challenge: the model, the application, and the user experience all matter.
What's next for CricketIQ: AI-Powered Cricket Shot Recognition
My next goal is to improve prediction reliability, handle continuous cricket footage more effectively, and explore shot-level summaries that could help players and coaches review their performance.
In the future, CricketIQ could grow into a sports analytics platform that turns cricket footage into more useful, accessible insights.
CricketIQ is a step toward making cricket video analysis more intelligent, practical, and accessible through AI.
Log in or sign up for Devpost to join the conversation.