Skip to content

Latest commit

 

History

History
40 lines (24 loc) · 814 Bytes

README.rst

File metadata and controls

40 lines (24 loc) · 814 Bytes

Curt: Layout Analysis with Transformers

Curt is an experimental layout analysis system based on transformers. It uses the set prediction approach of DETR to directly predict baselines encoded as Bézier curves and line classes. There is also an additional instance segmentation head that predicts line bounding polygons.

Installation

$ pip install .

Training

To train on ALTO or Page XML files on a GPU:

$ curt train -d cuda train *.xml

Warning: Curt is resource-intensive and needs at least 30Gb of GPU memory to train.

To train the instance segmentation head from an existing curve prediction model:

$ curt polytrain -d cuda -i checkpoint.pth *.xml

Inference

To be implemented.