Multitask Learning of Height and Semantics From Aerial Images

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 .

Boundary-Aware Multitask Learning for Remote Sensing Imagery

“…Compared with the separate training of a single task, the MTL scheme improves the performance of both segmentation and height tasks, indicating that joint learning of related tasks can benefit each other to learn more generalized features. This is consistent with the conclusion of [11], [12]. Table V shows a smaller λ r may not work effectively to play its role, while a larger λ r may over-smooth the predictions of two master tasks.…”

“…In all experiments, we empirically set the weight for three [8], [58] and IMG2DSM [7] are proposed for the remote sensing purpose while the D3Net model [76] is for the general scenes. The models most closely related to our method is the joint learning scheme [11], [12] to accomplish segmentation and height tasks together, which are re-implemented following their configurations under the same experimental settings for a fair comparison.…”

“…For the height estimation task, we report 628 the results from the algorithms of [7], [8], [58], [76]. The 629 comparison experiments with the state-of-the-art joint learning 630 methods [11], [12] are also provided. We further analyze the 631 influence of different backbones by performing the ResNet50 632 and ResNet101 trunks within the encoder backbone.…”

Current Challenges in Operational Very High Resolution Land-Cover Mapping

“…In deep-based architectures, one can now couple networks that share computation and knowledge for the benefit of several tasks: multi-task learning. It is beneficial for instance for unseen image generation or data completion such as Digital Surface Model (Carvalho et al, 2019), and should now be extended to other modalities. A particular case, and a major trend in computer vision, is panoptic segmentation (Kirillov et al, 2019), which includes instance segmentation.…”

Investigating the Aerodynamic Surface Roughness Length over Baghdad City Utilizing Remote Sensing and GIS Techniques

“…The height resulted from the difference between the DSM and DEM which is known as Digital Height Model (DHM). It represents the height surface element (Z H ) (7).…”