Semantic Segmentation of Remote Sensing Images with Sparse Annotations

TL;DR

This paper introduces a novel framework for semantic segmentation of high-resolution aerial images using sparse scribble annotations, leveraging a new regularization method to improve learning from limited labeled data.

Contribution

It proposes FESTA, a regularization technique that enhances segmentation performance by exploiting neighborhood structures in spatial and feature spaces with sparse annotations.

Findings

FESTA improves segmentation accuracy with sparse annotations.

The framework reduces annotation effort significantly.

Results outperform baseline methods on aerial image datasets.

Abstract

Training Convolutional Neural Networks (CNNs) for very high resolution images requires a large quantity of high-quality pixel-level annotations, which is extremely labor- and time-consuming to produce. Moreover, professional photo interpreters might have to be involved for guaranteeing the correctness of annotations. To alleviate such a burden, we propose a framework for semantic segmentation of aerial images based on incomplete annotations, where annotators are asked to label a few pixels with easy-to-draw scribbles. To exploit these sparse scribbled annotations, we propose the FEature and Spatial relaTional regulArization (FESTA) method to complement the supervised task with an unsupervised learning signal that accounts for neighbourhood structures both in spatial and feature terms.