EfficientDet is a state-of-the-art object detection model for real-time object detection originally written in Tensorflow and Keras but now having implementations in PyTorch--this notebook uses the PyTorch implementation of EfficientDet. It has an EfficientNet backbone and a custom detection and classification network. Because of this backbone, EffcientDet is designed to efficiently scale from the smallest model size. The smallest EfficientDet, EfficientDet-D0 has 4 million weight parameters - it is truly tiny. EfficientDet infers in 30ms in this distribution and is considered and can be stored with only 17 megabytes of storage--making it both a small and fast model.
EfficientDet performed state-of-the-art on COCO when it was released and performs slightly better than YOLOv3.
Training EfficientDet with Custom Data: https://blog.roboflow.com/training-efficientdet-object-detection-model-with-a-custom-dataset/