Instructions for extraction of segmentation mask from pretrained Mask-RCNN model

Inference model is based on Resnet101 backbone and is pre-trained on MS COCO dataset.

Prerequisites

Code is intended to work with Python 3.6.x , it hasn't been tested with previous version. This code can run on CPU/GPU backend.

Inference

  pip install -r requirements.txt

  python3 mask_generate.py --image_dir ./images --out_dir ./masks

You can also specify which type of objects to segment out here object_list = ['car', 'truck'].

Acknowledgements

Code is heavily borrowed from this repository : https://github.com/matterport/Mask_RCNN

I merely created a simple wrapper around it.

Input

Output