ROOR

ROOR is a reading order prediction (ROP) benchmark which annotates layout reading order as ordering relations.

Layout reading order is typically formulated as a permutation of layout elements, i.e. a sequence containing all the layout elements. However, multiple cases have reflected that this formulation does not adequately convey the complete reading order information in the layout, which may negatively affect the utilization of this signal in downstream VrD tasks.

Therefore, this work investigate the properties of layout reading order, conceptualizing it with terms Immediate Succession During Reading(ISDR) and Generalized Succession During Reading(GSDR), and formulate each of which as an ordering relation over layout elements. Then, ROOR provides the annotation of the ISDR relationship over layout segments, based on the layout annotation of EC-FUNSD. Overall, ROOR comprises 199 samples including 10,662 segments, 31,297 words and 10,967 annotated reading order linking pairs. We hope the construction of this benchmark could facilitate the development of automated ROP methods of the improved task form.