Is the U-Net Directional-Relationship Aware?

TL;DR

This paper investigates whether standard U-Nets can learn and utilize directional relationships between objects in segmentation tasks, revealing their capacity for relational reasoning when trained with sufficient data and receptive field size.

Contribution

The study demonstrates that U-Nets can learn directional relationship reasoning in segmentation tasks, challenging assumptions about their contextual understanding capabilities.

Findings

U-Nets can learn directional relationships with enough data and receptive field size

The network's reasoning about directional relationships can be analyzed through perturbation scenarios

U-Nets exhibit relational reasoning abilities in segmentation tasks

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.