- The paper introduces the largest annotated cuneiform sign dataset and uses a Deformable Detection Transformer (DETR) for object detection.
- The model is evaluated on 173 and 106 class granularities, achieving 28-37% improvement over prior work on COCO-style metrics.
- The system integrates automatic tablet-side extraction, heuristic line grouping, and n-gram-based textual similarity for visual sign detection.
- Applied to 87,668 tablet fragments from the eBL corpus, the method produces nearly 2.9 million sign detections.
- The approach operates without linguistic priors but is sensitive to tablet damage and layout variability, providing a scalable foundation for cuneiform analysis.