What: Technical AI session

When: Thursday 14th of January from 10am to 12.30pm

For Who: This session is aimed at AI developers or people already working with AI in the geospatial sector.

Object detection using Deep Learning techniques is an emerging technology that is shaping the way many organizations are collecting, storing or updating data. In this context it is important to know the challenges associated with it especially in context to geospatial data: changing objectives, types of data and level of automation etc,. During this session, speakers with a multi-disciplinary background will present and discuss their work on Geo AI focussing on AI for Object detection. These technical presentations will be followed by discussions and questions

The program is hosted by the Dutch Cadastre. This session is closely connected with the “Proeftuin BGT change detection” project therefore, the topics are also closely related with the ongoing work within the Proeftuin that is focussing on automating Key basic registrations using AI.

 

Sign up!

 

Program

10.00 – 10.10    Welcome and introduction

10.10 – 10.40    Devis Tuia

10.40 – 11.10    Wufan Zhao

11.10 – 11.20    Break

11.20 – 11.50    Bertrand Le Saux

11.50 – 12.20    Sander Oude Elberink

12.20 – 12.30    Wrap up

 

Bertrand Le Saux

Senior scientist, European Space Agency, Italy.

Ground truth data has a major influence on the results of a model and more often than not ground truth data is not as good as we hope or expect. However, it is an integral part of training a model and has to be dealt with. How can we handle this problem best? and what are the most important lessons to create robust AI models?

#WeaklySupervisedLearning #Semi-supervisedLearning #ContinualLearning

 

Wufan Zhao

PhD candidate, Faculty of Geo-Information Science and Earth Observation University of Twente, Netherlands

Currently the focus is on detecting objects or changes. Can these objectives be achieved with guaranteed high quality? The next step will be not only to locate but also to automatically delineate or map these features.

#BuildingOutlineDelineation #DeepLearningForBuildingExtraction #SegmanticSegmentation #InstanceSegmentation

 

Devis Tuia

Associate professor, EPFL ENAC, Environmental Computational Science and Earth Observation Laboratory (ECEO), Switzerland

As an associate professor in the field of machine learning and deep learning, Devis is UpToDate on this topic. What are the current trends and which research is worthwhile following? Which papers recently caught his attention and why?

#MachineLearning #DeepLearning #Innovation #Research

 

Sander Oude Elberink

Assistant Professor, Faculty of Geo-Information Science and Earth Observation University of Twente, Netherlands

Sander has done a lot of work in the field of point clouds and 3D reconstruction. His recent work focusses on combining point clouds with map data, a difficult but promising combination which could help in identifying topographic changes.

#DataFusion #Pointclouds #ChangeDetection