TUM-DLR Summer School 2017

October 4–6, 2017, at the JUFA Hotel Wangen, Allgäu.

Date and Venue

This year's TUM-DLR Summer School will take place at the JUFA Hotel Wangen - Sport-Resort, close to Lake Constance.


Wednesday, October 4th
by 10:30Arrival at Wangen & Check-In
10:30Welcome & InstructionsCamp Nou
11:00Poster TeasersCamp Nou
14:00Poster Session 1Camp Nou
15:30Poster Session 2Camp Nou
17:00Keynote 1
Prof. Vincent Lepetit: Deep Learning for 3D Localization
Camp Nou
18:00Leisure time
Thursday, October 5th
9:00Keynote 2
Bertrand Le Saux: Semantic Labeling of Remote Sensing Data, from 2D to 3D
Camp Nou
10:30Workshop Session 1
Introduction to Python and Version Control with GITCamp Nou
Exploring Generative Adversarial NetworksWembley
14:30Group Discussions
15:00Social Event
Friday, October 6th
9:00Research "Meet & Greet"Camp Nou
& Wembley
10:30Workshop Session 2
Google Earth EngineCamp Nou
Exploring the Potential of Geo-tagged Social Media Data in GeoscienceWembley
14:30Professor/Supervisor InputCamp Nou
15:30Closing and FeedbackCamp Nou
16:00Check-Out & Departure

This agenda is tentative and subject to change. You can import our Google Calendar (https://calendar.google.com/calendar/ical/surq4e6lor2j86lh0t49vafqrc%40group.calendar.google.com/public/basic.ics) in order to stay up-to-date.

Invited Speakers

Prof. Vincent Lepetit

Computer Vision for Augmented Reality Lab, TU Graz, Austria
Laboratoire Bordelais de Recherche en Informatique, U Bordeaux, France
Inria Manao, France

Deep Learning for 3D Localization

The first part of the talk will describe a novel method for 3D object detection and pose estimation from color images only. We introduce a "holistic"’ approach that relies on a representation of a 3D pose suitable to Deep Networks and on a feedback loop. This approach, like many previous ones is however not sufficient for handling objects with an axis of rotational symmetry, as the pose of these objects is in fact ambiguous. We show how to relax this ambiguity with a combination of classification and regression. The second part will describe an approach bridging the gap between learning-based approaches and geometric approaches, for accurate and robust camera pose estimation in urban environments from single images and simple 2D maps.

Bertrand Le Saux, PhD

ONERA, The French Aerospace Lab, Palaiseau, France
École Polytechnique, Paris, France

Semantic Labeling of Remote Sensing Data, from 2D to 3D

This talk will be about scene understanding with neural networks. Precisely, it starts from a brief introduction about classification of aerial and satellite images and the advent of deep learning for solving this task, then discusses various kinds of deep networks for dense semantic classification, fusion of heterogeneous data (especially with residual correction), and joint-learning with additional cartography. In a second part, it moves to 3D with semantic labeling of point clouds and presents SnapNet a multi-view convnet which can classify 3D from LiDAR or photogrammetry. It discusses various strategies for urban modeling or robotic exploration. Building on latest developments of the last years, we will see how it is now possible to semantize the world that surrounds us.


Introduction to Python and Version Control with GIT

Organizers Lloyd Hughes, TUM Professorship of Signal Processing in Earth Observation
Time Thursday, October 5th, 10:30 – 13:00

Python is an open source and free alternative to Matlab. Almost all tasks that are currently done using Matlab can be implemented with Python as well. Whereas GIT helps scientists to track changes and facilitates collaboration in software development.

In this workshop, we will introduce scientific Python stack. Students are invited (after the proposal has been accepted) to suggest areas of interest they would like to learn more about. Furthermore, an introduction into the general operation of GIT will be given.

Exploring Generative Adversarial Networks

Organizers Sandra Aigner, TUM Chair of Remote Sensing Technology
Jian Kang, TUM Professorship of Signal Processing in Earth Observation
Time Thursday, October 5th, 10:30 – 13:00
Description Generative Adversarial Networks (GANs) generate data that resembles the structure of a given training dataset.Important applications of GANs are: dataset augmentation, image editing and image inpainting.

Exploring the Potential of Geo-tagged Social Media Data in Geoscience

Organizers Diao Lin, TUM Chair of Cartography
Ruoxin Zhu, TUM Chair of Cartography
Lichao Mou, DLR-SiPEO
Time Friday, October 6th, 10:30 – 13:00
Description The increasing geo-tagged social media data provides new perspectives for researchers to get insight to various aspects of our society, like human mobility, events, opinions, and living environments. Such crowd-sourced data also draw attention of researchers from Geoscience. Due to the big volume and high complexity of social media dataset, it’s quite challenging to process and utilize the geo-tagged social media data. This workshop will firstly give an overall introduction of geo-tagged social media data and provide some practical guidance on how to collect, store, and preprocess the twitter data (one typical social media data). Then, from the perspective of Cartography and Remote Sensing, three extra application demos will be presented. The first demo illustrates how to use the social media data to detect spatio-temporal events and how to visualize different kinds of events (i.e. event mapping). The second demo presents how to conduct sentiment analysis based on the social media. In the third demo, we explore the potential of the geo-tagged social media data, e.g., twitter data, for the semantic understanding in the urban area. We hope these three demos will help the participants to get a better understanding of how geo-tagged social media can be applied in Geoscience.

Google Earth Engine

Organizers Lukas Liebel, TUM Chair of Remote Sensing Technology
Tobias Koch, TUM Chair of Remote Sensing Technology
Chunping Qiu, TUM Professorship of Signal Processing in Earth Observation
Time Friday, October 6th, 10:30 – 13:00
Description Google Earth Engine combines a multi-petabyte catalog of satellite imagery and geospatial datasets with planetary-scale analysis capabilities and makes it available for scientists, researchers, and developers to detect changes, map trends, and quantify differences on the Earth's surface.

Social Event

On Thursday afternoon (4-7 pm), we booked 10 lanes at the

Seaside Bowling Center
Meistershofener Str. 14
88045 Friedrichshafen.

Afterwards, we will have a joint buffet-style dinner there.

Further Material

You can access further course material from our Google Drive folder.


For registration, please fill out this form.


Chair of Remote Sensing Technology
Professorship of Signal Processing in Earth Observation
Professorship of Photogrammetry and Remote Sensing
Chair of Cartography