DocLang: Multilingual Document Type Classification

A multilingual document type classification dataset for identifying various document and visual content types. The dataset contains 13,200 images across 11 languages, with 1,200 images per language.

Dataset Structure

The dataset is organized by language code:

• ar/ - Arabic (1,200 images)
• bg/ - Bulgarian (1,200 images)
• de/ - German (1,200 images)
• en/ - English (1,200 images)
• es/ - Spanish (1,200 images)
• fr/ - French (1,200 images)
• hi/ - Hindi (1,200 images)
• it/ - Italian (1,200 images)
• ja/ - Japanese (1,200 images)
• ru/ - Russian (1,200 images)
• zh/ - Chinese (1,200 images)

All images are in JPG format and represent the test split.

Task

This dataset is designed for image-based document type classification.