Old photographs from Journalistic Picture Archive JOKA

This is an image dataset consisting of 4595 old photographs (until 1940) from the collections of the Journalistic Picture Archive JOKA, obtained from the Finna.fi discovery service. The images are intended to be used for different AI/ML tasks such as generating captions or colorizing them.

The images themselves are JPEG files under the directory images.

The metadata.jsonl file contains metadata about each image, for example descriptive captions (mostly in Finnish), subject keywords, location and the name of the photographer. The file is in JSON Lines format (each line is a JSON record). The file_name column contains the relative path of the JPEG image file.

The Python script collect-joka.py was originally used to retrieve the metadata and images from the from Finna API.

The images have been published by the Finnish Heritage Agency in Finna using the Creative Commons Attribution 4.0 license and these attribution instructions:

The file can be used freely, even commercially. You must cite the name of the author and/or photographer as well as the name of the collection and organisation in connection with the file. Instructions for use: The user must respect copyright and privacy protection and ensure that any data is correct. Using portraits in marketing and advertising is prohibited without permission from the people pictured. The user is responsible for acquiring permits concerning the people, works and trademarks represented in the files. The user is responsible for data processing in accordance with the purpose of use. Please provide the Finnish Heritage Agency’s library with a complimentary copy of the printed publication in which the file has been used. If you require a file of better quality, please contact Picture Collections of the Finnish Heritage Agency (kuvakokoelmat@museovirasto.fi). Documents: The material was subjected to optical character recognition (OCR) during digitisation, meaning that you may perform searches of the typed lists within the digital material. Character recognition always contains a margin of error, which should be observed when searching within digitised documents. Character recognition may miss words due to the quality or typeface of the original material. 3D: A light version of the 3D model suitable for online presentations has been published in the FINNA service.