GridStemLayNet - A Document Layout Understanding Model

Document Layout Understanding and Analysis solves the real-time problem of observing and contextualizing Layout and textual characteristics of a document. Document pre-trained models have proven to be very effective on various tasks in Document AI. However, for the document layout analysis (DLA) task, existing document pre-trained models, even those pretrained in a multi-modal fashion, usually are large and rely on complex visual features and just try to segment or contextualize high-level layouts but does not capture low level textual features. This research proposes a model that encompasses both layout features and textual features. To fully leverage multi-modal information, it uses Global Invariance and exploits efficient training techniques to learn better representation for DLA and textual properties. This research proposes a vision transformer based approach with Stem Attention mechanism. For the purpose of this research a new dataset called DTFiL will also be introduced for this specific problem.