Awesome

PublayNet-maskrcnn-mlnet

Using a MaskRCNN model trained on the PublayNet dataset from phamquiluan/PubLayNet with ML.Net in C# / .Net for Document layout analysis and page segmmentation task.

Recognised regions/categories are:

• Text (i.e. paragraph)
• Title
• List
• Table
• Figure

About

• About the PublayNet dataset: ibm-aur-nlp/PubLayNet
• Original repos for training the model: phamquiluan/PubLayNet
• Original MaskRCNN-Resnet50-FPN checkpoint here
• Jupyter notebook to convert for ONNX model are here
• Compressed ONNX model in PublayNetModelTEst/Assets/Model

Results

See here