Terms of Use

In order to use this database, you must firstly agree to the terms.
You may not use the database if you don't accept the terms.

The use of this database is limited to scientific and non-commercial purpose only, in the computer science domain.

For instance, you are allowed to split the images, through the use of segmentation algorithms.
You can also use pieces of this database to illustrate your research in publications and presentations.
Any other use case must be validated by our service.

If you do agree to be bound by all of these Terms of Use, please fill and send this request form using your academic email address to and then use the login and password provided to download the selected version below.

In case of published paper using this database, please notify us by email so we can reference your paper in the reference section of this website.
For any citation of this database, please use the BibTeX below:

Citation BibTeX

title = {eBDtheque: a representative database of comics},
author = {Gu{\'e}rin, Cl{\'e}ment and Rigaud, Christophe and Mercier, Antoine and Ammar-Boudjelal, Farid and Bertet, Karell and Bouju, Alain and Burie, Jean-Christophe and Louis, Georges and Ogier, Jean-Marc and Revel, Arnaud},
keywords = {scientific comics collection, ground truth, semantic annotation, image, database},
booktitle = {Proceedings of the 12th International Conference on Document Analysis and Recognition (ICDAR)},
year = {2013}


Version 3 - July 2019

  • 100 pages
  • 850 panels
  • 1081 balloons
  • 1620 comic characters
  • 4693 text lines
- add missing comic character and balloon associations
- fix two wrong comic character / balloon associations
- fix one Japanese character transcription
- replace negative polygon point coordinates by zero

Click here to download (complete request form first)

Version 2 - June 2014

  • 100 pages
  • 850 panels
  • 1092 balloons
  • 1550 comic characters
  • 4691 text lines
- Japanese text transcription
- Comic character locations
- Pixel level balloon segmentation
- Semantic relations between spoken text lines, speech balloons and speaking characters

Click here to download (complete request form first)

Version 1 - August 2013

  • 100 pages
  • 848 panels
  • 1091 balloons
  • 4667 text lines
- Japanese text transcription are missing (files INOUE_KYOUMEN_0xx.svg)
- French accents and cedilla may be not transcribed properly (e.g. É, È, À, Ç, é, è, à, ç transcribed as E, A, C, e, a, c)
- The metadata "queueDirection" has been renamed to "tailDirection" to be more explicit

Click here to download (complete request form first)


- add idPanel attribute (parent) to balloon and comic character polygon tags
- add names to comic characters (+ unamed)
- add comic character segmentation masks