Geo-Spatiotemporal Features and Shape-Based Prior Knowledge for Fine-grained Imbalanced Data Classification

Charles A. Kantor (1, 2, 3, 4) , Marta Skreta (1, 5) , Brice Rauby (2, 6) , Léonard Boussioux (7, 2) , Emmanuel Jehanno (2, 6) , Alexandra Luccioni (1, 8) , David Rolnick (1, 9) , Hugues Talbot (2, 6, 3, 10)

1 MILA - Montreal Institute for Learning Algorithms [Montréal]
2 CentraleSupélec
3 OPIS - OPtimisation Imagerie et Santé
4 Stanford University
5 University of Toronto
6 CVN - Centre de vision numérique
7 MIT - Massachusetts Institute of Technology
8 UdeM - Université de Montréal
9 McGill University = Université McGill [Montréal, Canada]
10 Université Paris-Saclay

Fine-grained classification aims at distinguishing between items with similar global perception and patterns, but that differ by minute details. Our primary challenges come from both small inter-class variations and large intra-class variations. In this article, we propose to combine several innovations to improve fine-grained classification within the use-case of wildlife, which is of practical interest for experts. We utilize geo-spatiotemporal data to enrich the picture information and further improve the performance. We also investigate state-of-the-art methods for handling the imbalanced data issue.

Mots clés: Deep Learning; Computer Vision; Machine Learning; Artificial Intelligence AI

Domaines: Machine Learning [stat.ML]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Biodiversité et Ecologie; Ecologie, Environnement; Apprentissage [cs.LG]; Intelligence artificielle [cs.AI]; Bio-informatique [q-bio.QM]; Réseau de neurones [cs.NE]

Charles A. Kantor, Marta Skreta, Brice Rauby, Léonard Boussioux, Emmanuel Jehanno, et al.. Geo-Spatiotemporal Features and Shape-Based Prior Knowledge for Fine-grained Imbalanced Data Classification. IJCAI 2021 - Workshop on AI for Social Good, Jan 2021, Tokyo, Japan. ⟨hal-03150686⟩