Weakly-supervised continual learning for class-incremental segmentation

TL;DR

This paper introduces a weakly-supervised continual learning method for class-incremental segmentation in remote sensing, addressing background shift and forgetting by regularization and pseudo-labeling, validated on public datasets.

Contribution

It proposes a novel approach combining regularization and pseudo-labeling to enable efficient class-incremental segmentation with weak supervision.

Findings

Effective on three remote sensing datasets

Reduces catastrophic forgetting

Handles background shift in continual learning

Abstract

Transfer learning is a powerful way to adapt existing deep learning models to new emerging use-cases in remote sensing. Starting from a neural network already trained for semantic segmentation, we propose to modify its label space to swiftly adapt it to new classes under weak supervision. To alleviate the background shift and the catastrophic forgetting problems inherent to this form of continual learning, we compare different regularization terms and leverage a pseudo-label strategy. We experimentally show the relevance of our approach on three public remote sensing datasets. Code is open-source and released in this repository: https://github.com/alteia-ai/ICSS}{https://github.com/alteia-ai/ICSS.