Automated sign detection across the Electronic Babylonian Library: A large-scale dataset and end-to-end cuneiform OCR pipeline

A large-scale annotated cuneiform sign dataset has been developed alongside an end-to-end cuneiform OCR pipeline utilizing a Deformable Detection Transformer (DETR) model, evaluated with 173 and 106 classes. The system integrates automatic extraction, heuristic line grouping, and n-gram-based textual similarity evaluation, achieving 28-37% improvements on COCO-style detection metrics and processing 87,668 tablet fragments to yield nearly 2.9 million sign detections. This framework enhances the scalability and interpretability of cuneiform analysis, paving the way for future multimodal and linguistic modeling applications.