Construction et évaluation d'un corpus pour la recherche d'instances d'images muséales

Abstract

This paper presents two datasets of annotated photos and videos from two museums. The data comes from two different museums : the Musée de Grenoble, with mainly paintings, and the Lyon-Fourvière museum, with Celtic and pre-Roman objects. In total, they contain 4674 annotated images, corresponding to 784 different artworks, and 3h07 of museum visit firstperson videos shot by 5 persons. This dataset can be used as a challenge for image retrieval and video segmentation and annotation tasks. They are freely available to the research community. The images of these collections contain 361 queries on a corpus of 4313 documents. Moreover, 2132 additional images are extracted from the visit videos, allowing to test images from other sources. Three state of the art approaches are processed and tested on these collections. MOTS-CLÉS : Recherche d’instances images, corpus.