Core YOLO data pipeline and training framework.
Things not included yet:
- A Network with Weights Trained until Convergence
- Extensive Data Augmentation Features
- Metric to Compute mAP during and after Training
- An exact Network Rebuild of the YOLOv2 or YOLOv3 Network ( Because this project/framework leaves the exact network definition open to the user)
Things included:
- Scripts to Transform VOC-annotated Data to YOLO Format
- Functions to Load, Preprocess, and Partially Augment the Data in a Scalable Way
- Custom-Built YOLO Loss Functions to Minimize
- Functions to Extract Final Bounding Boxes from YOLO Output Tensors