Missouri Botanical Garden Open Conference Systems, TDWG 2016 ANNUAL CONFERENCE

Font Size: 
Large-scale Evaluation of Multimedia Analysis Techniques for the Monitoring of Biodiversity
Alexis Joly, Hervé Goeau, Pierre Bonnet

Building: CTEC
Room: Auditorium
Date: 2016-12-07 04:30 PM – 04:45 PM
Last modified: 2016-11-24

Abstract


Computer-assisted identification of living organisms is considered as one of the most promising solutions to help bridging the taxonomic gap and build accurate knowledge of the geographic distribution and evolution of species. LifeCLEF (www.lifeclef.org) is a worldscale research forum dedicated to the evaluation of multimedia-oriented identification systems. Its principle is to measure and boost the performance of the state-of-the-art by sharing large-scale experimental data covering thousands of species. Each year, hundreds of research groups specialized in computer vision, audio processing, machine learning or data management register to the proposed challenges. Tens of them succeed in processing the whole data and submit technical papers describing their running system. Results are then synthetized and further analysed in joint research papers. The LifeCLEF research platform is globally organized around 3 tasks related to multimedia information retrieval and fine-grained classification problems in 3 subdomains. Each task is based on large and collaboratively revised data and the measured challenges are defined in collaboration with biologists and environmental stakeholders in order to reflect realistic usage scenarios.

The first task deals with image-based plant identification and is organized since 2011. It is based on a growing collaborative data collection produced by tens of thousands of members of a French social network of amateur and expert botanists. In 2015, this dataset contained 113,205 pictures of herb, tree and fern specimens belonging to 1,000 species (living in France and neighbouring countries). The second task deals with audio-based bird identification and is based on the audio recordings collected by a very active nature watchers network called Xeno-canto (http://www.xeno-canto.org/). This web-oriented community of bird sound recordists accounts for about 2,000 contributors that have already collected more than 180,000 recordings of about 9,000 species. Dataset used for the BirdCLEF task is focused on more than 20,000 audio recordings belonging to the 1000 bird species represented in the South-American region. The last task deals with the identification of sea organisms in general, from fish to whales to dolphins to sea beds to corals.

In this talk, we will report the main outcomes of the 2016-th edition of LifeCLEF including a comprehensive description of the best performing methods.  We will then discuss perspectives of future developments according to the growing available datasets, and interest of the scientific community for this lab.