Title: Simple Games for Complex Classification ProblemsMaximilien Servajean
03/07/2015, 11h
Abstract:
Pl@ntNet is a large-scale innovative participatory sensing platform relying on image-based plants identification as a mean to enlist non-expert contributors and facilitate the production of botanical observation data. The iOs and Android mobile applications allowing to identify plants and share observations have been downloaded by more than 350K users in 170 countries and counts up to ten thousands users per day. Nowadays, the whole collection contains more than 180K images covering about 7K plant species (mainly in West Europe). The collected data, revised by a large social network specialized in botany (TelaBotanica), is used by several international projects including the Gbif initiative and the LifeCLEF evaluation campaign which involves hundreds of research groups working on automatic recognition tools and methods. However there is still a need for human validation and identification when the application failed to propose a correct identification, such cases happening especially for the rarest and most interesting plants. Crowdsourcing has shown a lots of interests in the recent years. In such approaches, users are ask to resolve micro-tasks which results are then aggregated using mathematical tools in order to create knowledge. Unfortunately, asking a large set of users to identify some random plants is merely impossible. In this presentation, I will show some preliminary works we are doing to combine automatic identification tools and crowdsourcing in order to identify the maximum possible number of plants.
