Train your own Table Structure Recognition model using Mask-RCNN

The original implementation of Matterport is here

It also has a super-resolution model to enhance the image quality before sending to OCR. Reference is from here

For OCR Google's Tesseract is used here.

The datasets are provided under dataset_custom and dataset_marmot

Additions:

• A Flask based UI for easy testing of the trained model.
• To test the model use extract_tables.ipynb

The pretrained weights are here. Put it in the models directory