Computer Vision
MLOps
YOLOv8
Knowledge Distillation
FastAPI
AWS
A lightweight, production-ready object detection pipeline designed for robotics, where latency, model size, and real-time performance matter most. The system uses knowledge distillation to transfer accuracy from a YOLOv8s teacher model into a compact YOLOv8n student — ideal for edge devices and constrained hardware.
What This Project Includes
- YOLOv8 teacher → student knowledge distillation
- End-to-end MLOps pipeline from training to deployment
- MLflow Model Registry for version control & experiment tracking
- Airflow orchestration for scheduled retraining & drift detection
- Real-time inference via FastAPI (with Prometheus latency metrics)
- AWS-based training and artifact storage using EC2 + S3
Tech Stack
Python, YOLOv8, PyTorch, MLflow, Apache Airflow, AWS EC2/S3, Docker, FastAPI, Prometheus.
Full details, metrics, and visual results are available in the project’s GitHub README.