YOLOv5-deeplearning-notebook

automated pipeline for training custom YOLOv5 object detection models using the YouTube Bounding Box dataset

• simple notebook interface
• supports 23 different object classes (person, cat, dog, vehicles, etc.)
• processes YouTube videos into training-ready datasets
• supports cloud or local GPU resources for training models

usage

1. set parameters in yolobook.ipynb
2. run cells sequentially -> get trained model + performance metrics
3. upload a photo or video to test model inference

automated pipeline features

data processing (src/utility/process-data.py):

• downloads YouTube videos (added multithreading, ~3x faster now)
• extracts frames efficiently (a single ffmpeg call per video)
• generates YOLO-format bounding box labels
• remaps class IDs to zero-indexed format
• splits data into train/validation/test sets
• handles cleanup and error cases

YOLOv5 integration:

• pre-configured for accepted dataset formats
• automated YAML config generation
• ready-to-run training pipeline

utilities:

• visual-check.py - randomized dataset inspection with bounding box overlays
• datapaths-to-txt.py - generates image path files for training