Study: Application of Mask R-CNN on Historical Document Image Dataset

This is an implementation of instance segmentation on a challenging datasets--historical document image dataset--based on Mask R-CNN on Python 3, Keras, and TensorFlow.

Requirements

To rerproduce our result, please consider whether your machine is installed with the following requirements:

• Python 3.4
• TensorFlow 1.3
• Keras 2.0.8
• Other common packages listed in requirements.txt.

Usage

To train a model, please run python [deep_voronoi or bolts] train --dataset=/path/to/dataset/ --model=[imagenet or coco]