DeepLabScan - Computer Vision Object Detection Pipeline
End-to-end ML pipeline for object detection using YOLO architectures with PyTorch, achieving 100% automation from raw data to model evaluation.
Developed an end-to-end ML pipeline for object detection using YOLO architectures (v8/v11) with PyTorch, achieving 100% automation from raw data to model evaluation.
Key Features
Intelligent Data Augmentation: System with geometric and photometric transformations, automatically balancing imbalanced datasets with configurable limits to prevent overfitting.
Modular Experiment Tracking: Excel-based system enabling comparison of 100+ experiments with COCO-style metrics (mAP@0.5:0.95, Precision, Recall, F1-Score).
Performance Optimization: Automatic hardware detection (CUDA/MPS/CPU), early stopping, and dynamic batch sizing, supporting real-time inference from webcam/video/images.
Technologies
PyTorch 2.9+, Ultralytics YOLO, OpenCV 4.12+, NumPy, Pandas, Matplotlib, Seaborn, scikit-learn. ~3,100 lines of production code.