A document is worth a structured record: Principled inductive bias design for document recognition

ArXi:2507.08458v2 Announce Type: replace-cross Many document types use intrinsic, convention-driven structures that serve to encode precise and structured information, such as the conventions governing engineering drawings. However, many state-of-the-art approaches treat document recognition as a mere computer vision problem, neglecting these underlying document-type-specific structural properties, making them dependent on sub-optimal heuristic post-processing and rendering many less frequent or complicated document types inaccessible to modern document recognition.