Multi-Task Learning of Height and Semantics from Aerial Images

TL;DR

This paper introduces a neural network framework for jointly learning land semantics and height from aerial images, improving accuracy and providing uncertainty estimates, demonstrated on a large dataset.

Contribution

The novel multi-task learning framework simultaneously predicts semantics and height, enhancing performance and offering uncertainty quantification.

Findings

Joint learning improves accuracy for both tasks.

Framework provides uncertainty maps for predictions.

Effective on large aerial imagery dataset.

Abstract

Aerial or satellite imagery is a great source for land surface analysis, which might yield land use maps or elevation models. In this investigation, we present a neural network framework for learning semantics and local height together. We show how this joint multi-task learning benefits to each task on the large dataset of the 2018 Data Fusion Contest. Moreover, our framework also yields an uncertainty map which allows assessing the prediction of the model. Code is available at https://github.com/marcelampc/mtl_aerial_images .