EARTHVISION 2020 – Large Scale Computer Vision for Remote Sensing Imagery

About

Earth observation and remote sensing are ever-growing fields of investigation where computer vision, machine learning, and signal/image processing meet. The general objective is to provide large-scale, homogeneous information about processes occurring at the surface of the Earth exploiting data collected by airborne and spaceborne sensors. This workshop aims at fostering collaboration between the computer vision and Earth Observation communities to boost automated interpretation of remotely sensed data and to raise awareness inside the vision community for this highly challenging and quickly evolving field of research with a big impact on human society, economy, industry, and the planet.

Find latest up to date information at the (external) EarthVision20 webpage or follow EV20 on Twitter:
 

Presentation Format

For regular papers, we follow the CVPR schedule: Each paper is presented by a pre-recorded video that is accessible throughout the WS. Questions and comments to the authors can be made via a text-based chat (also available throughout the WS) as well as during two live video sessions 12h apart.

The opening and closing sessions as well as most of the keynotes will be live, i.e. there is no prerecorded video but a live virtual session for the presentation as well as to interact with the presenters. The keynotes will be given only during the first round, i.e. there is no second live presentation in the second round after 12h. One keynote will be given as video with no Q&A session.

The SpaceNet 6 Challenge session has additional material on the external SN6 webpage.

Program (1st round, 2nd round +12h without keynote talks)

Find the corresponding google calendar here.

08:30 – 08:40 Welcome

08:45 – 09:25 Keynote 1 – Markus Reichstein, Max Planck Institute, Germany – (Live talk + Q&A, only in the first round)

09:30 – 10:30 Q&A Session 1 – Applied Deep Learning – (Video + live Q&A, both rounds)

10:30 – 11:00 Break

11:00 – 11:35 Keynote 2 – Sveinung Loekken, ESA – Title: The Contours of a trillion-pixel Digital Twin Earth – (Live talk + Q&A, only in the first round)

11:40 – 12:40 SpaceNet 6 Challenge: www.spacenet.ai/earthvision2020

12:40 – 14:00 Break

14:00 – 14:30 Keynote 3 – Serge Belongie, Cornell Tech., USA – Title: A New Workflow for Collaborative Machine Learning-Research in Biodiversity – (Live talk + Q&A, only in the first round)

14:35 – 15:45 Q&A Session 2 – Synthesis and Semantic Analysis

15:45 – 16:00 Break

16:00 – 16:40 Keynote 4 – Maros Blaha, Apple – Title: Towards Large-Scale Remote Sensing of Human Habitats – (Video, no Q&A)

16:40 – 17:00 Awards & Closing – (Live, both rounds)

Teaser picture for paper
The EarthVision workshop aims at fostering collaboration between the CV and EO communities to boost automated interpretation of remotely sensed data.
    Authors: Introductory remarks   
    Keywords:  Earth Observation, Remote Sensing, Machine Learning, Computer Vision
Sund Jun14  
8:30 AM - 8:40 AM
Favorite
Teaser picture for paper
Markus Reichstein will address how to combine artificial intelligence with system modelling for enhanced insight into the complex Earth System.
    Authors: Markus Reichstein   
    Keywords:  climate, terrestrial vegetation, soil
Sund Jun14  
8:45 AM - 9:25 AM
Favorite
Teaser picture for paper
Instead of learning stereo reconstruction from scratch, we train a deep residual network to refine an initial height map with the help of stereo image
    Authors: Corinne Stucker, Konrad Schindler   
    Keywords:  learned residual stereo reconstruction, urban DEM refinement, building priors, satellite imagery
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
Monitoring active regions in the farside of the Sun is important for space weather forecasting. However, direct imaging of the farside is currently no
    Authors: Rasha Alshehhi   
    Keywords:  Generative Adversarial Network, Multi-scale, Active Region, Solar Magnetogram, Atmospheric Imaging Assembly, Extreme UltraViolet Imager
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
We present an automated approach, RasterNet, for estimating free-flow driving speed by fusing overhead imagery and aerial LiDAR point clouds using a g
    Authors: Armin Hadzic, Hunter Blanton, Weilian Song, Mei Chen, Scott Workman, Nathan Jacobs   
    Keywords:  LiDAR, point cloud, urban understanding, image-driven mapping, transportation engineering, remote sensing
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
We present DALES a large-scale semantic segmentation data set for aerial LiDAR. DALES contains over half a billion hand labeled point, spanning 10 squ
    Authors: Nina Varney, Vijayan K Asari, Quinn Graehling   
    Keywords:  LiDAR, LADAR, Point cloud, Semantic segmentation, Aerial, Airborne, ALS, Data Set, 3D, Laser, Aerial vision, Airborne system, Earth scan, Deep learning, Data annotation, Object category
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
Toronto-3D: A Large-scale Mobile LiDAR Dataset for Semantic Segmentation of Urban Roadways. Dataset available at https://github.com/WeikaiTan/Toronto-
    Authors: Weikai Tan, Li Jonathan   
    Keywords:  Mobile LiDAR, dataset, point cloud, semantic segmentation, deep learning
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
Sen1Floods11 introduces a flood specific dataset to train and validate flood detection from Sentinel-1 SAR imagery. Here we show Sentinel-1 and Sentin
    Authors: Derrick J Bonafilia, Tyler Anderson, Erica Issenberg, Beth Tellman   
    Keywords:  flooding, convolutional neural networks, open dataset, Sentinel-1, Sentinel-2, water detection
Sund Jun14  
9:30 AM - 10:30 AM
Favorite
Teaser picture for paper
What might a dynamic, 4D, science- and observation-based high resolution digital twin of the earth look like?
    Authors: Sveinung Loekken   
    Keywords:  Digital Twins, Earth Observation, Remote Sensing, Artificial intelligence, AI4EO, Big data, Phi-lab, Copernicus
Sund Jun14  
11:00 AM - 11:35 AM
Favorite
Teaser picture for paper
The SpaceNet 6 challenge asked participants to automatically extract building footprints from a combination of SAR and optical satellite imagery.

    Keywords:  Synthetic Aperture Radar (SAR), open data, multi-sensor data, building footprints, benchmark, semantic segmentation
Sund Jun14  
11:40 AM - 12:40 PM
Favorite
Teaser picture for paper
A new workflow for biodiversity research institutions who would like to make use of Machine Learning based on billion+ species occurrence count.
    Authors: Serge Belongie   
    Keywords:  biodiversity, GBIF, mediated data
Sund Jun14  
2:00 PM - 2:30 PM
Favorite
Teaser picture for paper
In this paper, we propose a Density-Map guided object detection Network (DMNet) for object detection in aerial images. DMNet has three key components:
    Authors: Changlin Li, Taojiannan Yang, Sijie Zhu, Chen Chen, Shanyue Guan   
    Keywords:  object detection in aerial images, small objects, deep learning, density estimation
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Top and bottom rows depict the original and the standardized satellite images, respectively. StandardGAN learns to standardize multiple images by taki
    Authors: Onur Tasar, Yuliya Tarabalka, Alain Giros, Pierre Alliez, Sébastien Clerc   
    Keywords:  Generative adversarial networks, GANs, Image standardization, Semantic segmentation, Multi-source domain adaptation, Style transfer
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Qualitative results of our FGCN: Semantic Segmentation architecture on Semantic3D benchmark dataset.
    Authors: Saqib Ali Khan, Yilei Shi, Muhammad Shehzad, Xiaoxiang Zhu   
    Keywords:  semantic segmentation, graph convolutional networks, feature extraction, state-of-the-art, open source
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Creating models that generalize across space has been difficult in Earth observation, because remotely sensed data experience distributional shifts fr
    Authors: Marc Rußwurm, Sherrie Wang, Marco Körner, David Lobell   
    Keywords:  Meta-Learning, Few-Shot, Transfer Learning, Land Cover Classification, Remote Sensing, Earth Observation
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Source domain consists of existing high resolution bands: NIR(R), R(G), G(B). Attention map corresponds to the spatial attention coefficients extra
    Authors: Litu Rout, Indranil Misra, Manthira Moorthi S, Debajyoti Dhar   
    Keywords:  Spatial Attention, Laplacian Spectral Attention, Super Resolution, Multi-Spectral Band Synthesis, Wasserstein GAN
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Qualitative analysis on WorldView-2. SPOA outperforms compared methods in perceptual quality, and also generates more natural textures while mitigatin
    Authors: Litu Rout, Saumyaa Shah, Manthira Moorthi S, Debajyoti Dhar   
    Keywords:  Monte-Carlo Sampling, Reinforcement Learning, Siamese Policy, Super Resolution, Remote Sensing.
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
High-resolution satellite imagery is critical for various earth observation applications related to environment monitoring, geoscience, forecasting, a
    Authors: Md Rifat Arefin, Vincent Michalski, Pierre-Luc St-Charles, Alfredo Kalaitzis, Sookyung Kim, Samira Ebrahimi Kahou, Yoshua Bengio   
    Keywords:  Multi Image Super Resolution, Satellite Imagery, Gated Recurrent Unit, Convolutional Neural Network
Sund Jun14  
2:35 PM - 3:45 PM
Favorite
Teaser picture for paper
Urbanization spurs the demand for scalable transportation solutions, capable of navigating densely populated areas and enabling smooth mobility.
    Authors: Maros Blaha   
    Keywords:  Urbanization, transportation solutions, navigation, mobility
Sund Jun14  
4:00 PM - 4:40 PM
Favorite
Teaser picture for paper
The closing session will not only look back on the challenges of having EV20 as a virtual event but also announce the winners of the Best Paper Award.

    Keywords:  Earth Observation, Remote Sensing, Machine Learning, Computer Vision
Sund Jun14  
4:40 PM - 5:00 PM
Favorite
Teaser picture for paper
The EarthVision workshop aims at fostering collaboration between the CV and EO communities to boost automated interpretation of remotely sensed data.
    Authors: Introductory remarks   
    Keywords:  Earth Observation, Remote Sensing, Machine Learning, Computer Vision
Sund Jun14  
8:30 PM - 8:40 PM
Favorite
Teaser picture for paper
Instead of learning stereo reconstruction from scratch, we train a deep residual network to refine an initial height map with the help of stereo image
    Authors: Corinne Stucker, Konrad Schindler   
    Keywords:  learned residual stereo reconstruction, urban DEM refinement, building priors, satellite imagery
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
Monitoring active regions in the farside of the Sun is important for space weather forecasting. However, direct imaging of the farside is currently no
    Authors: Rasha Alshehhi   
    Keywords:  Generative Adversarial Network, Multi-scale, Active Region, Solar Magnetogram, Atmospheric Imaging Assembly, Extreme UltraViolet Imager
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
We present an automated approach, RasterNet, for estimating free-flow driving speed by fusing overhead imagery and aerial LiDAR point clouds using a g
    Authors: Armin Hadzic, Hunter Blanton, Weilian Song, Mei Chen, Scott Workman, Nathan Jacobs   
    Keywords:  LiDAR, point cloud, urban understanding, image-driven mapping, transportation engineering, remote sensing
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
We present DALES a large-scale semantic segmentation data set for aerial LiDAR. DALES contains over half a billion hand labeled point, spanning 10 squ
    Authors: Nina Varney, Vijayan K Asari, Quinn Graehling   
    Keywords:  LiDAR, LADAR, Point cloud, Semantic segmentation, Aerial, Airborne, ALS, Data Set, 3D, Laser, Aerial vision, Airborne system, Earth scan, Deep learning, Data annotation, Object category
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
Toronto-3D: A Large-scale Mobile LiDAR Dataset for Semantic Segmentation of Urban Roadways. Dataset available at https://github.com/WeikaiTan/Toronto-
    Authors: Weikai Tan, Li Jonathan   
    Keywords:  Mobile LiDAR, dataset, point cloud, semantic segmentation, deep learning
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
Sen1Floods11 introduces a flood specific dataset to train and validate flood detection from Sentinel-1 SAR imagery. Here we show Sentinel-1 and Sentin
    Authors: Derrick J Bonafilia, Tyler Anderson, Erica Issenberg, Beth Tellman   
    Keywords:  flooding, convolutional neural networks, open dataset, Sentinel-1, Sentinel-2, water detection
Sund Jun14  
9:30 PM - 10:30 PM
Favorite
Teaser picture for paper
The SpaceNet 6 challenge asked participants to automatically extract building footprints from a combination of SAR and optical satellite imagery.

    Keywords:  Synthetic Aperture Radar (SAR), open data, multi-sensor data, building footprints, benchmark, semantic segmentation
Sund Jun14  
June 14 - June 15
Favorite
Teaser picture for paper
In this paper, we propose a Density-Map guided object detection Network (DMNet) for object detection in aerial images. DMNet has three key components:
    Authors: Changlin Li, Taojiannan Yang, Sijie Zhu, Chen Chen, Shanyue Guan   
    Keywords:  object detection in aerial images, small objects, deep learning, density estimation
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
Top and bottom rows depict the original and the standardized satellite images, respectively. StandardGAN learns to standardize multiple images by taki
    Authors: Onur Tasar, Yuliya Tarabalka, Alain Giros, Pierre Alliez, Sébastien Clerc   
    Keywords:  Generative adversarial networks, GANs, Image standardization, Semantic segmentation, Multi-source domain adaptation, Style transfer
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
Qualitative results of our FGCN: Semantic Segmentation architecture on Semantic3D benchmark dataset.
    Authors: Saqib Ali Khan, Yilei Shi, Muhammad Shehzad, Xiaoxiang Zhu   
    Keywords:  semantic segmentation, graph convolutional networks, feature extraction, state-of-the-art, open source
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
Creating models that generalize across space has been difficult in Earth observation, because remotely sensed data experience distributional shifts fr
    Authors: Marc Rußwurm, Sherrie Wang, Marco Körner, David Lobell   
    Keywords:  Meta-Learning, Few-Shot, Transfer Learning, Land Cover Classification, Remote Sensing, Earth Observation
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
Source domain consists of existing high resolution bands: NIR(R), R(G), G(B). Attention map corresponds to the spatial attention coefficients extra
    Authors: Litu Rout, Indranil Misra, Manthira Moorthi S, Debajyoti Dhar   
    Keywords:  Spatial Attention, Laplacian Spectral Attention, Super Resolution, Multi-Spectral Band Synthesis, Wasserstein GAN
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
Qualitative analysis on WorldView-2. SPOA outperforms compared methods in perceptual quality, and also generates more natural textures while mitigatin
    Authors: Litu Rout, Saumyaa Shah, Manthira Moorthi S, Debajyoti Dhar   
    Keywords:  Monte-Carlo Sampling, Reinforcement Learning, Siamese Policy, Super Resolution, Remote Sensing.
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
High-resolution satellite imagery is critical for various earth observation applications related to environment monitoring, geoscience, forecasting, a
    Authors: Md Rifat Arefin, Vincent Michalski, Pierre-Luc St-Charles, Alfredo Kalaitzis, Sookyung Kim, Samira Ebrahimi Kahou, Yoshua Bengio   
    Keywords:  Multi Image Super Resolution, Satellite Imagery, Gated Recurrent Unit, Convolutional Neural Network
Mond Jun15  
2:35 AM - 3:45 AM
Favorite
Teaser picture for paper
The closing session will not only look back on the challenges of having EV20 as a virtual event but also announce the winners of the Best Paper Award.

    Keywords:  Earth Observation, Remote Sensing, Machine Learning, Computer Vision
Mond Jun15  
4:40 AM - 5:00 AM
Favorite

 

Sponsors – Gold Level

           

Sponsors – Bronze Level

Sponsors – Technical