Plateforme Camomile d'annotation collaborative de documents multimédia
Project : Camomile Polymer Client
Web page : https://github.com/camomile-project/camomile-polymer-client
Source coding : https://github.com/camomile-project/camomile-polymer-client
Axis : DataSense
Coordinator : Bredin Hervé
Candidate : BEAUMONT Romain
Administrator Laboratory: LIMSI
Team : TLP
Engagement : 01/11/2016 —> 30/06/2017
The Camomile collaborative annotation platform for multimedia documents (image, audio, video and text) was developed as part of the CHISTERA eponymous project (www.chistera.eu/projects/camomile) which expired in 2016. It meets the growing need for manually annotated data (speech and language processing, computer vision, etc.) in different scientific domains of the DataSense axis.
Based mainly on Node.js, MongoDB and Docker technologies, the platform provides a REST API that facilitates the production of interfaces (mobile or web) for creating, editing and sharing multimedia document annotations (audio, video, image, text).
In 2015 and 2016, the Camomile platform was used to support the organization of an evaluation campaign on the identification of people in television streams (tinyurl.com/zyq5cal).
The CHISTERA JOKER (www.chistera.eu/projects/joker) and FUI GUIMUTEIC (www.guimuteic.fr/) projects have also chosen the platform to annotate their own video data. In each of these different cases of use, one of the developers of the platform was involved in some way in the implementation of the annotation tool.
Despite the availability of documentation and both clients, a developer seeking to design his own annotation interface based on the Camomile platform will face a steep learning curve. It is therefore a matter of designing a starter kit providing a simple and modular web interface interacting with the Camomile platform, as well as educational documentation.
Expected Results :
The purpose of this starter kit is to increase the distribution of the Camomile platform, by increasing the number of cases of use of the platform, which has hitherto been used mainly for annotation purposes for the automatic processing of audiovisual sequences. In particular, the natural language processing community could greatly benefit from such an annotation platform for its text corpus. We also anticipate its use in the humanities and social sciences where the collaborative aspect of the platform could facilitate its use in a "crowdsourcing" context.
As for Camomile, this funding made it possible to perpetuate the work carried out in the eponymous project and to relaunch the annotation activities which had been on forced pause since the end of the initial CHISTERA project.