FedDrive v2: an Analysis of the Impact of Label Skewness in Federated Semantic Segmentation for Autonomous Driving

TL;DR

FedDrive v2 investigates how label skewness in federated learning impacts semantic segmentation performance in autonomous driving, introducing new scenarios and analyzing the role of domain information during testing.

Contribution

The paper introduces six federated scenarios to study label skewness effects and compares them with domain shift, enhancing understanding of federated semantic segmentation challenges.

Findings

Label skewness significantly affects model performance.

Domain information during testing can mitigate skewness effects.

New scenarios provide insights into federated learning dynamics.

Abstract

We propose FedDrive v2, an extension of the Federated Learning benchmark for Semantic Segmentation in Autonomous Driving. While the first version aims at studying the effect of domain shift of the visual features across clients, in this work, we focus on the distribution skewness of the labels. We propose six new federated scenarios to investigate how label skewness affects the performance of segmentation models and compare it with the effect of domain shift. Finally, we study the impact of using the domain information during testing. Official website: https://feddrive.github.io