Tech Stack
Object Detection
YOLO
Faster R-CNN
Model Optimization
PyTorch
Description
Compared Faster R-CNN and YOLOv8 object detection frameworks to evaluate performance trade-offs between two-stage and single-stage architectures for real-time applications.
Analyzed architectural complexity versus computational efficiency to identify optimal deployment strategies for production computer vision systems.
- Achieved 96.1% mAP@50 with YOLOv8n single-stage detector
- Reduced training time by 3.2x (47 minutes vs 2.5 hours) while maintaining superior accuracy
- Evaluated comprehensive metrics: 87.73% mAP@75 and 71.12% mAP@50-95
- Implemented Region Proposal Networks with SGD optimization for Faster R-CNN baseline