SpaceNet 6: Multi-Sensor All Weather Mapping Dataset

Shermeyer, Jacob; Hogan, Daniel J.; Brown, John J.; Etten, Adam Van; Weir, N.; Pacifici, Fabio; Hänsch, Ronny; Bastidas, Alexei; Soenen, Scott; Bacastow, Todd; Lewis, Ryan

doi:10.1109/cvprw50498.2020.00106

Cited by 94 publications

(53 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The other reason is that most of the buildings in these three cities are moderate size and height. As point out in [60], the size and height of buildings influence the performance. Smaller buildings could not be detected, which exist in the datasets of Shanghai, Beijing, Rio, and Yokosuka.…”

Section: Results and Analysis A Investigation On A Single Modelmentioning

confidence: 96%

“…Fourth, we investigated the performance with different pre-training weights, including Imagenet, Instagram, SSL on Imagenet, SWSL on Imagenet, from the encoder of ResNeXt101 32×8d. We also consider the transfer learning approach [60]. In this case, the model is first trained on RGB datasets, and then the generated weights are used as the initial weights for training on SAR.…”

Section: Results and Analysis A Investigation On A Single Modelmentioning

confidence: 99%

See 1 more Smart Citation

A Benchmark High-Resolution GaoFen-3 SAR Dataset for Building Semantic Segmentation

Xia

Yokoya

Adriano

et al. 2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Deep learning is increasingly popular in remote sensing communities and already successful in land cover classification and semantic segmentation. However, most studies are limited to the utilization of optical datasets. Despite few attempts applied to synthetic aperture radar (SAR) using deep learning, the huge potential, especially for the very high-resolution SAR, are still underexploited. Taking building segmentation as an example, the very high resolution (VHR) SAR datasets are still missing to the best of our knowledge. A comparable baseline for SAR building segmentation does not exist, and which segmentation method is more suitable for SAR image is poorly understood. This paper first provides a benchmark high-resolution (1 m) GaoFen-3 SAR datasets, which cover nine cities from seven countries, review the state-of-the-art semantic segmentation methods applied to SAR, and then summarize the potential operations to improve the performance. With these comprehensive assessments, we hope to provide the recommendation and roadmap for future SAR semantic segmentation.

show abstract

Section: Results and Analysis A Investigation On A Single Modelmentioning

confidence: 96%

Section: Results and Analysis A Investigation On A Single Modelmentioning

confidence: 99%

A Benchmark High-Resolution GaoFen-3 SAR Dataset for Building Semantic Segmentation

Xia

Yokoya

Adriano

et al. 2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…To address this issue, Wang et al [67] take building polygons from the OpenStreetMap (OSM) data set and an official map as ground truth data and train a network to segment buildings in an urban scene. For building footprint extraction, Shermeyer et al [68] presented a multisensor all-weather mapping (MSAW) data set containing airborne SAR images, optical images, and building footprint annotations, along with a deep network baseline model and benchmark. However, in these two works, building footprints, instead of building areas, are learning targets.…”

Section: B Related Workmentioning

confidence: 99%

CG-Net: Conditional GIS-Aware Network for Individual Building Segmentation in VHR SAR Images

Sun

Hua

Mou

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Object retrieval and reconstruction from very-highresolution (VHR) synthetic aperture radar (SAR) images are of great importance for urban SAR applications, yet highly challenging due to the complexity of SAR data. This article addresses the issue of individual building segmentation from a single VHR SAR image in large-scale urban areas. To achieve this, we introduce building footprints from geographic information system (GIS) data as a complementary information and propose a novel conditional GIS-aware network (CG-Net). The proposed model learns multilevel visual features and employs building footprints to normalize the features for predicting building masks in the SAR image. We validate our method using a high-resolution spotlight TerraSAR-X image collected over Berlin. Experimental results show that the proposed CG-Net effectively brings improvements with variant backbones. We further compare two representations of building footprints, namely, complete building footprints and sensor-visible footprint segments, for our task, and conclude that the use of the former leads to better segmentation results. Moreover, we investigate the impact of inaccurate GIS data on our CG-Net, and this study shows that CG-Net is robust against positioning errors in the GIS data. In addition, we propose an approach of ground truth generation of buildings from an accurate digital elevation model (DEM), which can be used to generate large-scale SAR image data sets. The segmentation results can be applied to reconstruct 3-D building models at level-of-detail (LoD) 1, which is demonstrated in our experiments.

show abstract

“…By pretraining on comprehensive image datasets such as ImageNet [6], multiple works have achieved state of the art results and/or faster convergence for many segmentation tasks [28,11,10,15], including the SpaceNet baseline model [29]. Recent work [29,15] has also shown that improvements can be gained by pretraining on datasets more closely aligned to the task at hand, such as the Carvana dataset [1]. In attempts to leverage multiple satellite views, the SpaceNet baseline also pretrains the networks on the more informative EO satellite views.…”

Section: Related Workmentioning

confidence: 99%

“…This is a realistic assumption inspired from real-world settings, where as aforementioned, EO data are likely to be missing due to conditions at capture time. The straightforward approach towards such a challenging setting is to pretrain a network with the view that is missing at test time, and subsequently fine-tune the network using data available both during training and testingan approach commonly employed in satellite imagery analysis [28,10,29,1]. At the same time, several approaches have been proposed on multi-view learning for satellite imagery for a variety of tasks [22,18,19].…”

Section: Introductionmentioning

confidence: 99%

Shared-Space Autoencoders with Randomized Skip Connections for Building Footprint Detection with Missing Views

Ashiotis

Oldfield

Chrysostomou

et al. 2021

Pattern Recognition. ICPR International Workshops and Challenges

View full text Add to dashboard Cite

Recently, a vast amount of satellite data has become available, going beyond standard optical (EO) data to other forms such as synthetic aperture radars (SAR). While more robust, SAR data are often more difficult to interpret, can be of lower resolution, and require intense pre-processing compared to EO data. On the other hand, while more interpretable, EO data often fail under unfavourable lighting, weather, or cloud-cover conditions. To leverage the advantages of both domains, we present a novel autoencoder-based architecture that is able to both (i) fuse multi-spectral optical and radar data in a common shared-space, and (ii) perform image segmentation for building footprint detection under the assumption that one of the data modalities is missing-resembling a situation often encountered under real-world settings. To do so, a novel randomized skip-connection architecture that utilizes autoencoder weight-sharing is designed. We compare the proposed method to baseline approaches relying on network fine-tuning, and established architectures such as UNet. Qualitative and quantitative results show the merits of the proposed method, that outperforms all compared techniques for the task-at-hand.

show abstract

SpaceNet 6: Multi-Sensor All Weather Mapping Dataset

Cited by 94 publications

References 24 publications

A Benchmark High-Resolution GaoFen-3 SAR Dataset for Building Semantic Segmentation

A Benchmark High-Resolution GaoFen-3 SAR Dataset for Building Semantic Segmentation

CG-Net: Conditional GIS-Aware Network for Individual Building Segmentation in VHR SAR Images

Shared-Space Autoencoders with Randomized Skip Connections for Building Footprint Detection with Missing Views

Contact Info

Product

Resources

About