This is the initial release of `tf2_yolo`, which aims to provide a solid TensorFlow 2.x implementation of YOLOv4: Optimal Speed and Accuracy of Object Detection.
This initial release contains the code for inference of YOLOv4 on the CSPDarknet53 backbone. Weights are converted from the original Darknet implementation.
It includes:
- [x] Original YOLOv4 implementation (CSPDarknet53, SPP, YOLOv3 Head)
- [x] Inference using the Keras functional API
- [x] A CLI command to convert the original Darknet weights to a compatible .h5 format
- [x] Normalized outputs, so that bounding boxes can be plotted on the original input (instead of the resized one)