CuneiML: A Cuneiform Dataset

This is the data presented in the article CuneiML: A Cuneiform Dataset for Machine Learning, published in the Journal of Open Humanities Data. Our dataset provides metadata and bounding boxes for 38,947 photographs of cuneiform tablets, as well as transliteration and cuneiform Unicode. The structure of the JSON file is the following:

{
 'id': 131837,
 'img_url': 'https://cdli.mpiwg-berlin.mpg.de/dl/photo/P131837.jpg',    # link to photo
 'lineart': 'https://cdli.mpiwg-berlin.mpg.de/dl/lineart/P131837_l.jpg',# link to lineart
 # the bounding box, [[x_1, y_1], [x_2, y_2]]
 # where (x_1, y_1) is the left upper vertex and (x_2, y_2) is the lower right vertex of the bounding box of the cutout
 'bboxes': [[204.0, 200.0], [523.0, 522.0]],                            
 'text': {
   'obverse': [
      {'raw': '2(gesz2) 4(asz) 2(barig) 4(disz) sila3 gur',
         'num': '1',
         'sign': ['𒐂', '𒐉', '𒋡', '𒄥']},
      {'raw': 'a2 lu2 hun-ga2', 
         'num': '2', 
         'sign': ['𒀉', '𒇽', '𒂠', '𒂷']},
      {'raw': 'ugu2 lu2- <D> inanna ba-a-gar',
         'num': '3',
         'sign': ['𒀀𒅗', '𒇽', '<D>', '𒈹', '𒁀', '𒀀', '𒃻']}],
   'reverse': [
      {'raw': 'mu sza-asz-ru ki  ba-hul',
       'num': '1',
       'sign': ['𒈬', '𒊭', '𒀸', '𒊒', '𒆠', '𒁀', '𒅆𒌨']}]
   }
'geo': 'Umma (mod. Tell Jokha)',
'time': 'Ur III (ca. 2100-2000 BC)',
'genre': 'Administrative',
}

For more information on the dataset, visit our GitHub repository.