mask-rcnn-wrapper

This project provides a simple GUI wrapper around the Mask R-CNN implementation by matterport.

The wrapper provides an interface allowing you to pick a video to apply the model to. It will apply the model frame by frame while displaying the output and optionally saving the output to a new video file.

While the video is rendering you can enable/disable object detection (and sub options such as the masks and bounding boxes) in real time allowing you to apply the model to only certain portions of the video at a time.

Setup

Weights

Create a new folder inside the root project directory called weights, download the weight file:

1. Go to https://github.com/matterport/Mask_RCNN/releases
2. Download mask_rcnn_coco.h5
3. Place the file inside the weights folder

Direct download link.

Dependencies

cv2, pyqt5, numpy, scipy, cython, h5py, Pillow, scikit-image, tensorflow-gpu, keras, pycocotools

To install pycocotools:

1. git clone https://github.com/philferriere/cocoapi.git
2. pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

pycocotools itself requires Visual C++ 2015 Build Tools.

This project was developed and tested using Python 3.6

Credits

matterport - Mask R-CNN implementation

markjay4k - displaying frames with OpenCV instead of matplotlib