---
license: mit
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: val
    path: data/val-*
dataset_info:
  features:
  - name: page_arkindex_id
    dtype: string
  - name: page_image
    dtype: image
  - name: page_index
    dtype: int64
  - name: zone_orders
    sequence: int64
  - name: zone_polygons
    sequence:
      sequence:
        sequence: float64
  - name: zone_classes
    sequence:
      class_label:
        names:
          '0': HEADER-TITLE
          '1': HEADER-TEXT
          '2': SECTION-TITLE
          '3': ILLUSTRATION
          '4': ADVERTISEMENT
          '5': ANNOUNCEMENT
          '6': TITLE
          '7': TEXT
          '8': SUBTITLE
          '9': INSIDEHEADING
          '10': CAPTION
          '11': AUTHOR
          '12': TABLE
          '13': ILLUSTRATEDTEXT
          '14': TABLECONTENT
          '15': ASIDE
  - name: zone_texts
    sequence: string
  - name: zone_article_ids
    sequence: int64
  - name: zone_section_ids
    sequence: int64
  splits:
  - name: train
    num_bytes: 9252698919.392
    num_examples: 7957
  - name: test
    num_bytes: 498089523
    num_examples: 433
  - name: val
    num_bytes: 520680343
    num_examples: 446
  download_size: 10051632298
  dataset_size: 10271468785.392
task_categories:
- image-to-text
- text-classification
- image-classification
- image-segmentation
language:
- fr
tags:
- newspapers
pretty_name: Newspaper Finlam La Liberté
---

# Newspaper dataset: Finlam La Liberté

## Dataset Description

- **Homepage:** [Finlam](https://finlam.projets.litislab.fr/)
- **Point of Contact:** [TEKLIA](https://teklia.com)

## Dataset Summary

The **Finlam La Liberté** dataset includes 1500 issues from La Liberté, a French newspaper, from 1925 to 1928. 

Each issue contains multiple pages, with one image for each page resized to a fixed height of 2500 pixels.

The dataset can be used to train end-to-end newspaper understanding models, with tasks including:
* Text zone detection and classification
* Reading order detection
* Article separation

### Split

| set   | images | newspapers |
| ----- | ------:| ----------:|
| train | 7957   |  1350      |
| val   |  446   |   75       |
| test  |  433   |   75       |

### Languages

All newspapers in the dataset are French.

## Dataset Structure

### Data Fields

- `page_arkindex_id`: The Arkindex element id corresponding to the current page.
- `page_index`: The index of the current page in the newspaper issue.
- `page_image`: a PIL.Image object containing the page image. Note that when accessing the image column (using dataset[0]["page_image"]), the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["page_image"] should always be preferred over dataset["page_image"][0].
- `zone_polygons`: the list of zone coordinates in the current page.
- `zone_texts`: the list of zone texts in the current page.
- `zone_classes`: the list of zone classes in the current page.
- `zone_orders`: the list of zone indexes in the current page, defining the reading order.
- `zone_article_id`: the list of article indexes defining in which article of the newspaper the current zone is located.
- `zone_section_id`: the list of section indexes defining in which section of the newspaper the current zone is located.