Table detection has been an interesting and challenging problem in the field of document analyses. Object detection algorithms such as Faster-RCNN has been exploited so many times to detect tables in the documents. This is just another experiment but with a different architecture which is Mask_RCNN.
NOTE: I have transofrmed the Mask_RCNN repository from Matterport to do this research : https://github.com/matterport/Mask_RCNN