Heritage
Digitized photographic prints from Victor Forbin's personal archives, preserved by the Service historique de la Défense within their original archival organization.
Historical photographs · Document analysis · Digital humanities
The Forbin Dataset
More than 62,000 historical photographs from the archives of Victor Forbin, enriched with archival metadata, transcriptions, annotated stamps and automatic stamp detection.
Original archives held byService historique de la Défense
62,135historical photographs
120,106images, recto and verso
256archive boxes
CC BY-NC 4.0distribution license
The collection
Victor Forbin and his news picture agency
Victor Forbin began as a journalist in the 1880s, writing for Excelsior, the Journal des Voyages, the Revue des deux mondes and L’Illustration. In the early twentieth century he set up his own news picture agency: he bought prints in London, Berlin, Rome and New York and sold them to the main French newspapers. Its activity peaked in the 1910s and 1920s.
- Boxes256 digitized, sorted by geographical area, about 240 photographs each
- EnvelopesPhotographs grouped by theme
- PhotographsEach one digitized on both sides, recto and verso
One collection, three perspectives
From historical documents to research data
Research
Searchable metadata, places, subjects, captions and press-agency stamps for history, the history of photography and digital humanities.
Computational data
COCO polygons, OCR output and stamp detection predictions for historical document analysis and computer vision.
Research tasks
A benchmark for historical document analysis
Stamp detection
Photo agency and archive stamps are annotated on the verso of the prints, with model predictions for stamp detection and recognition.
OCR and handwriting
Typed and handwritten captions, titles, dates and signatures are located and transcribed for text recognition on historical documents.
Recto–verso linking
Each photograph pairs the image side with its annotated back, linking visual content with its archival and editorial history.
Captioning and similarity search
Paired images, texts and metadata support caption generation, image similarity search and metadata enrichment.
Circulation in the press
Visual similarity and archival metadata help trace how photographs were reused and republished.
Degradation analysis
Faded ink, stains and bleed-through offer a realistic test of robustness for recognition models.
Digital research desk
Study images, their context and their circulation
The explorer preserves each document's archival path, from its box to its recto or verso. Full-collection images are loaded on demand from Huma-Num Sharedocs using JSON manifests.
- Search metadata and transcriptions
- Compare annotations and predictions
- Zoom into recto and verso details
- Reuse structured data for research
Documentation
The dataset at a glance
- Archive holder
- Service historique de la Défense, series GR 2 K 247
- Content
- 62,135 photographs, 120,106 images (recto and verso)
- Digitization
- By Azentis, at 300 dpi, with Canon 5DSR and Fujifilm GFX 100 S cameras
- Metadata
- Class, continent, subject or country and sizes; 674 original labels grouped into 30 themes (in French)
- Formats
- JPEG images and COCO-style JSON
- File names
SHDGR__GR_2_K_247_157_010__0001.jpg: box 157, photograph 010,0001recto,0002verso- Languages
- French and English
- License
- CC BY-NC 4.0
- Published
- 9 December 2025
- Dataset DOI
- 10.57967/hf/7189
Citation
Cite the article or the dataset
The collection combines images, archival metadata and annotations designed for reproducible research.
Article
Read the publication · DOI
@article{Chelali2026_forbin,
author = {Chelali, Mohamed and Gosselet, Sylvain-Karl and Cloppet, Florence and Kurtz, Camille and Bloch, Isabelle and Foliard, Daniel},
doi = {10.5334/johd.487},
journal = {Journal of Open Humanities Data},
title = {The FORBIN Dataset: A Collection of Historical Photographs With Archival Metadata},
year = {2026}
}
Dataset
Open the dataset · DOI
@misc{Chelali2025_forbin_dataset,
author = {Chelali, Mohamed and Gosselet, Sylvain-Karl and Cloppet, Florence and Kurtz, Camille and Bloch, Isabelle and Foliard, Daniel},
title = {Forbin Dataset},
year = {2025},
publisher = {Hugging Face},
doi = {10.57967/hf/7189},
url = {https://doi.org/10.57967/hf/7189}
}About
HIGH VISION — Computer vision for a history of early news photo agencies
The Forbin Dataset is produced within the ANR HIGH VISION project (ANR-24-CE38-4079), which brings together historians and computer vision researchers to study the birth of an international visual culture through the archives of early news photo agencies: Forbin, Rol, Bain News and Black Star.
Dataset created by Mohamed Chelali, Sylvain-Karl Gosselet and Daniel Foliard (ECHELLES), Florence Cloppet and Camille Kurtz (LIPADE), and Isabelle Bloch (LIP6). With thanks to everyone who contributed to annotation and to identifying the circulation of photographs, in particular Sihem Yousfi and Marie-Louise Kitoko.
Archive holder
Service historique de la Défense
The original photographs of Victor Forbin are preserved by the Service historique de la Défense (SHD), under the reference GR 2 K 247. The SHD holds these archives; the images published here are digitized reproductions made available for research.
Visit the SHD website
Research consortium
ANR HIGH VISION
The dataset is produced by the HIGH VISION consortium, which brings together computer science laboratories, historians and heritage institutions.
Visit the HIGH VISION website
Funded by the French National Research Agency under grant ANR-24-CE38-4079.