Automatic species identification from photographs is central to modern biodiversity monitoring, but current operational systems (Pl@ntNet, iNaturalist, Merlin Photo ID) rely on black-box deep learning models that lack interpretable internal structure and degrade sharply on rare, previously unseen, or out-of-distribution species. Human experts, by contrast, identify unfamiliar specimens through explicit reasoning over morphological traits: structured, interpretable descriptors such as leaf shape, beak curvature, or wing pattern.
eTaxonomist is an ANR JCJC project (2026 to 2030) that aims to close this gap by developing computer vision methods that emulate expert, trait-based reasoning. The project consists of three work packages: constructing structured trait knowledge bases from expert sources (WP1), grounding this structured knowledge visually in images (WP2), and integrating both into an interpretable, zero-shot reasoning framework (WP3). The approach will be validated across three case studies of increasing taxonomic breadth: agriculturally important insects of France, birds, and plants worldwide.
The project will be under the supervision of Diego Marcos (Inria), Alexis Joly (Inria, Pl@ntNet co-founder) and Zeynep Akata (TU Munich) and will count with the support of expert taxonomists accross all taxonomic groups and with the Pl@ntNet platform.
