This repo is a simplification of the Mask-RCNN-TF2 by Ahmed Gad with some minor changes in the files:

1. mrcnn/model.py
2. mrcnn/utils.py

Mask-RCNN-TF2 extends the original Mask_RCNN project by incorporating TensorFlow 2.0. Use this repo or Mask-RCNN-TF2 with TF2 only as TF1 functionality is not supported.

This project has been tested using:

• tensorflow 2.2.0
• keras 2.3.1
• h5py 2.10

This project does not require installation. To run this successfully it is recommended to uninstall existing versions of TF, Keras, and H5PY as follows:

!pip uninstall keras -y
!pip uninstall keras-nightly -y
!pip uninstall keras-Preprocessing -y
!pip uninstall keras-vis -y
!pip uninstall tensorflow -y

!pip install tensorflow==2.2.0
!pip install keras==2.3.1
!pip install h5py==2.10.0

Copy the mrcnn directory into your project and you are good to go.

You will need the COCO weights so make sure you get a copy from here.

An example of using this project in Google Colab notebooks can be found here (coming shortly)

A great thank you to both team Matterport for this amazing piece of work, and to Ahmed Gad for making the effort of bring the original project up to date.

Use this bibtex to cite the original Matterport repository:

@misc{matterport_maskrcnn_2017,
  title={Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow},
  author={Waleed Abdulla},
  year={2017},
  publisher={Github},
  journal={GitHub repository},
  howpublished={\url{https://github.com/matterport/Mask_RCNN}},
}