Vision

The Vision-based auto-tagger. Uses document layout segmentation, reading-order detection and table-structure recognition to build a higher-fidelity structure tree. Does not generate figure descriptions.