ICDAR 2019 cBAD: 3,021 Annotated Historical Document Images

A dataset of 3,021 annotated document page images collected from seven European archives for the ICDAR 2019 Competition on Baseline Detection (cBAD). All baselines were manually annotated, and the training and evaluation sets include PAGE XMLs with annotated text regions and baselines. The dataset was created by Markus Diem of TU Wien.

Use Cases

• Benchmarking baseline detection algorithms based on manually annotated baselines.
• Training models for text region segmentation based on PAGE XML annotations.
• Researching document analysis techniques on a real-world collection of historical images.

Strengths

• Contains 3,021 annotated document page images.
• All baselines were manually annotated, suggesting high-quality ground truth.
• Includes training, evaluation, and test splits for competition benchmarking.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to its collection from seven European archives.

Provenance

Source

Seven European archives.

Collection Method

Images were collected and manually annotated.

Geography

Europe