Work

Is the U-Net Directional-Relationship Aware?

Year: 2022

Type: preprint

Abstract: CNNs are often assumed to be capable of using contextual information about distinct objects (such as their directional relations) inside their receptive field. However, the nature and limits of this capacity has never been explored in full. We explore a specific type of relationship~-- directional~-- using a standard U-Net trained to optimize a cross-entropy loss function for segmentation. We train this network on a pretext segmentation task requiring directional relation reasoning for success and state that, with enough data and a sufficiently large receptive field, it succeeds to learn the proposed task. We further explore what the network has learned by analysing scenarios where the directional relationships are perturbed, and show that the network has learned to reason using these relationships.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2207.02574

Authors: Riva, Mateus, Gori, Pietro, Yger, Florian, Bloch, Isabelle

Institutions: Télécom Paris, Institut Polytechnique de Paris, Laboratoire Traitement et Communication de l’Information, Université Paris Sciences et Lettres, Lamsade, Université Paris Dauphine-PSL, Centre National de la Recherche Scientifique, Sorbonne Université, LIP6

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funders: Agence Nationale de la Recherche, Fundação de Amparo à Pesquisa do Estado de São Paulo

Awards: G3010527862, Spatio-temporal analysis of pediatric magnetic resonance images, Intermediate representations in Computational Science for knowledge discovery, G4113406716, G4402799499, PaRis Artificial Intelligence Research InstitutE, G6803667113, Spatio temporal analysis of pediatric magnetic resonance images

OpenAlex ID: W4308093312

Cites: 17

Cited by:

Related to: 10