Google has been always at the forefront when it comes to artificial intelligence research. Google AI team has recently open-sourced ‘EfficientDet’, an advanced object detection tool using minimum compute. EfficientDet achieves better performance in comparison with YOLO or AmoebaNet when used with CPU or GPU’S.
EfficientDet is the next-generation version of EfficientNet which use to be one of the advanced object detection models released in early 2019 for Coral boards. EfficientDets are developed based on the advanced backbone, a new BiFPN, and a new scaling technique:
- Backbone: The more advanced EfficientNets are used in backbone networks.
- BiFPN: BiFPN is being proposed as a new bi-directional feature network to enable easy and fast feature fusion.
- Scaling: Proposed use of a single compound scaling factor to govern the network depth, width, and resolution for all backbone, feature network, and prediction networks.
Paper Pdf: https://arxiv.org/pdf/1911.09070.pdf
Github Codes: https://github.com/google/automl/tree/master/efficientdet
EfficientNet Tutorial: https://cloud.google.com/tpu/docs/tutorials/efficientnet