ModelDrivenSegmentation

NN based segmentation processor

Idea

Literature

• dhSegment: A generic deep-learning approach for document segmentation: Oliveira & Seguin, 2019

Existing tools

• https://github.com/ocr-d-modul-2-segmentierung/page-segmentation
• https://github.com/dhlab-epfl/dhSegment
• https://github.com/mjenckel/LAYoutERkennung

Related tutorials or blog posts

• https://fairyonice.github.io/Learn-about-Fully-Convolutional-Networks-for-semantic-segmentation.html

Model

Interface/Packaging

Maybe the actual work should be done in a separate component reading a (binarized/raw) image and writing a (pixel-classified) image, which merely gets wrapped for OCR-D interfaces and PAGE-XML GT.