My Projects

This project involved building a fightcam model that can accurately detect human actions and analyze actions directly from camera input with high accuracy. The system was trained on a dataset of 50,000 product reviews and achieved 89% accuracy on the test set.

Key features include:

• Real-time pose and face landmark detection using OpenCV & MediaPipe
• Action recognition pipeline to detect fights and aggressive gestures
• Integrated with live camera feed for instant predictions and alerts

A convolutional neural network designed to classify images into 10 different categories. The model was trained on ****** dataset and achieved **% accuracy through careful architecture design and data augmentation.

Technical highlights:

• Custom ResNet architecture with skip connections
• Implemented data augmentation pipeline to improve generalization
• Model pruning and quantization for mobile deployment