Mask-RCNN-TF2-3.0

The MaskRCNN class is replaced with 2 classes:

1. MaskRCNNDirected: For detecting the objects based on either the user-defined region proposals or the region proposals generated by the RPN.
2. MaskRCNNDirectedRPN: For returning the outputs of the region proposal layer. It either returns the region proposals generated by the RPN or [for error checking] returning the user-defined region proposals.

The decision of whether the region proposals are user-defined or generated by the RPN is made based on the REGION_PROPOSALS attribute in the mrcnn_directed.config.Config class. If this attribute is None, then the region proposals generated by the RPN are used. Otherwise, the user-defined region proposals are used.

In addition to the ProposalLayer, another layer is supported:

1. ProposalLayerDirected: Used when the user feeds a pre-defined region proposals.

In the mrcnn_directed.config.Config class, there is an attribute called REGION_PROPOSALS. If it is None, then the ProposalLayer is used. Otherwise, the ProposalLayerDirected is used.

Mask-RCNN-TF2-1.0

Supporting TensorFlow 2.0

Mask-RCNN-TF2-2.0

Support of a new model called Directed Mask R-CNN where the user directs the model to search only within some specific region.