Sign in to view source links and access this dataset
Description
The Finlam dataset contains 149 French newspapers from the 19th and 20th centuries. It includes 721 page images across training, validation, and test splits, with each page annotated for zones containing polygon, text, class, and order information. The dataset was created by Teklia and last updated on February 17, 2025.
Use Cases
Train document layout segmentation models based on annotated page zones.
Perform optical character recognition (OCR) on historical French text based on the provided text zones.
Analyze the evolution of newspaper design and structure over time based on the 19th-20th century date range.
Develop multimodal models that combine visual layout and textual content based on the image and text annotations.
Strengths
Contains 149 distinct historical newspapers, providing a substantial corpus.
Includes 721 annotated page images across three data splits (train, val, test).
Page images are standardized to a fixed height of 2000 pixels, which may aid model processing.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
The description is incomplete, referring users to an external page for full details.
Provenance
Source
Teklia
Collection Method
Likely digitized and annotated from historical newspaper archives.
Time Range
19th to 20th centuries
Freshness
Last updated 2025-02-17 08:58:55; freshness should be verified.
Geography
France (inferred from language and newspaper origin)
License is unknown; terms of use must be verified before download.