TexBiG (from the German Text-Bild-Gefüge, meaning Text-Image-Structure) is a document layout analysis dataset for historical documents in the late 19th and early 20th century. The dataset provides instance segmentation (bounding boxes and polygons/masks) annotations for 19 different classes with more then 52.000 instances. Annotations are manually annotated by experts and evaluated with Krippendorff's Alpha, for each document image are least two different annotators have labeled the document. The dataset uses the common COCO-JSON format.

TexBiG (from the German Text-Bild-Gefüge, meaning Text-Image-Structure) is a document layout analysis dataset for historical documents in the late 19th and early 20th century. The dataset provides instance segmentation (bounding boxes and polygons/masks) annotations for 19 different classes with more then 52.000 instances. Annotations are manually annotated by experts and evaluated with Krippendorff's Alpha, for each document image are least two different annotators have labeled the document. The dataset uses the common COCO-JSON format.

This version of the combines the test and train set to extend the training set, but leaves the validation set untouched. Also, the format was changed a bit to be easier to use. The original version of the dataset that was published alongside the publication at the GCPR (German Conference on Pattern Recognition 2022) can be found at zenodo.

TexBiG (from the German Text-Bild-Gefüge, meaning Text-Image-Structure) is a document layout analysis dataset for historical documents in the late 19th and early 20th century. The dataset provides instance segmentation (bounding boxes and polygons/masks) annotations for 19 different classes. Annotations are manually annotated by experts, with each image annotated five times. This annotations are then aggregated to approximate a ground truth.

This version of the dataset contains the test data, which are available on EvalAI for public evaluation.