Datasets:
scta
/

Modalities:
Image
Text
Formats:
parquet
Languages:
Latin
Libraries:
Datasets
Dask
jeffreycwitt's picture
updated readme
041af51 verified
|
raw
history blame
742 Bytes
metadata
dataset_info:
  features:
    - name: image
      dtype: string
    - name: bounding_boxes
      sequence:
        sequence: int32
    - name: transcriptions
      sequence: string
  splits:
    - name: train
      num_bytes: 17737371
      num_examples: 3006
  download_size: 8546225
  dataset_size: 17737371
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - la

Guiding Principles of this Dataset

Segmentation

  • The goal is here to segment the "main text" while ignoring paratext. This means page numbers, catch words, marginal annotations, are purposefully not segmented

Transcription

  • The guiding transcription principles are to
    • expand abbreviations
    • preserve orthography