Deep Learning Models for Table Detection in PDF Document images:
- RetinaNet (Working Model: Train and Test Functions for the model in RetinaNet.py)
- FasterRCNN (Archived Model: Train and Test Functions for the model in FasterRCNN.py)
- YOLOv3 (Archived Model)
Table Detection requires pre-processing of input image which is using distance transform and saving information provided by EuclideanDistanceTransform, LinearDistanceTransform, MaxDistanceTransform as three channels of the image. Method present in DetectTablesUtils.py as preProcessSampleImages(). Loopkup the sample files in data folder for the original pdf image and its distance transformed version.
Deep Learning Framework: Keras with Tensorflow
Dataset: For links to dataset of document images with tables, refer the enclosed Object Detection Details and Survey
SampleResults: Contains document images with tables detected by the RetinaNet model