Pipeline that loads data from Roboflow, augments it, and trains a YOLOv5 model on Vertex AI.
Pipeline that loads the trained model, builds a BentoML bundle, and deploys it to Vertex AI.
Pipeline that loads test data and runs predictions using the deployed model.
This project demonstrates how AI can bridge communication gaps for the deaf community by automatically recognizing American Sign Language (ASL) alphabet signs in real-time images. Using computer vision and modern MLOps practices, I've built an end-to-end pipeline that can detect and interpret ASL signs with high accuracy.
The system uses YOLOv5, a state-of-the-art object detection algorithm, to identify and classify hand signs representing the ASL alphabet. This enables real-time translation of sign language into text, making communication more accessible for deaf individuals.
The project leverages ZenML to orchestrate a sophisticated machine learning workflow:
- Data Processing: Automatically downloads and prepares ASL alphabet images from Roboflow
- Data Augmentation: Enhances training data using Albumentations to improve model robustness
- GPU-Accelerated Training: Trains the YOLOv5 model on Google Vertex AI with GPU support
- Experiment Tracking: Records all training metrics and parameters with MLflow
- Deployment: Packages the model with BentoML for production-ready inference
- Inference Pipeline: Provides a streamlined way to make predictions on new images
This project showcases how modern MLOps practices can be applied to create AI solutions that make a real difference in people's lives, while demonstrating advanced skills in computer vision, cloud computing, and machine learning engineering.
