Built an AI-powered cricket legality classifier using a pre-trained VGG19 model to detect legal vs. no-ball deliveries from bowler images, achieving 93% accuracy in real-world scenarios.
- Secured 1st place at Kshitij 2020 and won an INR 20,000 award for innovation and high model performance, outperforming peers in both accuracy and practical application.
- Fine-tuned a deep convolutional network on delivery images, with a focused pipeline for data preparation, augmentation and evaluation.