README.md

metadata

language:
  - en
dataset_info:
  features:
    - name: filename
      dtype: string
    - name: line_id
      list: string
    - name: line_text
      list: string
    - name: line_bboxes
      list:
        list: int64
    - name: word_id
      list: string
    - name: word_text
      list: string
    - name: word_bboxes
      list:
        list: int64
  splits:
    - name: train
      num_bytes: 8374521
      num_examples: 1539
  download_size: 2198686
  dataset_size: 8374521
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

IAM Dataset (Corrected & Reformatted)

A corrected and structured version of the IAM Handwriting Database, for text detection and text recognition tasks. This dataset offers segmentation for both line and word level.

Data fields

• filename: name of the image file
• line_id: list of line identifiers
• line_text: text list containing the text of each line
• line_bboxes: list of bounding boxes of each line (x, y, X, Y)
• word_id: list of word identifiers
• word_text: text list containing the text of each word
• word_bboxes: list of bounding boxes of each word (x, y, X, Y)

Each field is sorted from top-left to bottom-right.

For more information visit the official page