Fiche de SUN Rémy

Titre : Content combination strategies for Image Classification" (anglais) ou "Stratégies de combinaison de contenus pour la classification d’images
Date : 17 octobre 2022
Heure: 14:00
Résumé : In this thesis, we tackle the question of deep image classification, a fundamental issue for computer vision and visual understanding in general. Faced with neural networks' need for large training datasets, we show how mixing information from multiple images can be leveraged to improve the performance of the models. In a first part, we propose a new type of data augmentation that helps model generalize by embedding the semantic content of samples into the non-semantic context of other samples in a semi-supervised context. In a second part, we show input mixing can be used as an input compression method to train multiple subnetworks in a base network from compressed inputs. In a third part, we adapt this MMO technique to the emerging Vision Transformer (ViT) models. Our work shows ViTs present unique challenges for MIMO training, but that they are also uniquely suited for it.
Lieu : SCAI, Sorbonne Université