Complex Layout Classification in the Wild: A Low-Resource Approach with Layout-Preserving Augmentations
Researchers have developed a new method for classifying complex document layouts in low-resource scenarios. The approach utilizes a Convolutional Neural Network (CNN) combined with novel data augmentation techniques, including narrow anisotropic Gaussian masking and reflection-induced label transformations. These methods help the model learn global geometric arrangements by suppressing incidental text details while preserving essential structural information. The proposed strategy significantly improves page-level layout classification accuracy, even with severe annotation scarcity. AI
IMPACT This research offers a potential solution for improving document analysis in under-resourced languages or complex historical documents.