MSc defense by Larissa Gorzawski

On Wednesday 5th of October at 2pm Larissa will present her M.Sc. thesis “Deep transfer learning on street-level imagery for classification of seismic building types in Lima, Peru”.

From the abstract: “Comprehensive exposure models for seismic risk assessment require accurate building inventories in the endangered areas. The learning capabilities of Deep Learning (DL) can be combined with street-level imagery to categorize buildings in an automated way. Since the training of a DL model requires large amounts of data, in this thesis, a transfer learning approach will be employed to adapt an already trained model to a new study area and a different image dataset while minimizing the labeling requirements. The used model was trained by Aravena Pelizari, et al. (2021) with Google Street View (GSV) images in Santiago de Chile and will be adapted to the Peruvian capital Lima and to street-level imagery of the open-source platform Mapillary.

Three data-driven active learning (AL) strategies are designed and implemented with a pre-labeled pool of images: an initial cluster-based sampling with subsequent margin sampling and two additional augmentation strategies exploring the impact of either the most similar source domain images or secondly the addition of high-confidence semi-labeled target domain images.

The methods could achieve an improvement of the F1 accuracy score from 0.31 to 0.67 with a comparatively small amount of labeled images. Though the methods did all per-form similarly, the initial clustering and the semi-labeling of additional target domain images were the most promising approaches. The class accuracies indicate that the class differences between the domains could be learned at least partly quite successfully, while the performance of the data-driven AL methods was presumably limited by the noisiness of the dataset. The methods provided promising first results and could be further improved with diversity-based batch sampling as well as an extension of the semi-supervised learning approach.”

supervisors: Hannes Taubenböck and Christian Geiss

read more news:

EAGLES rock(et) the World Space Forum 2024

EAGLES rock(et) the World Space Forum 2024

Laura - 8th gen EAGLE and in the last week of her internship at UN-SPIDER - and Sunniva - 7th gen EAGLE and working as a student assistant at the DLR's space agency - together happen to help organize the World Space Forum (WSF) 2024 in Bonn. Surrounded by...

EAGLE internship at CIAT in Colombia

EAGLE internship at CIAT in Colombia

Leonie, an 8th generation EAGLE, is currently doing her internship at CIAT (International Center of Tropical Agriculture) in South America, Colombia. She is part of the Multifuncional Landscapes group, which investigates about soil organic carbon sequestration in...

Blender GIS introduction

Blender GIS introduction

Within out EAGLE Earth Observation M.Sc. we also cover software applications which might not be used on a regular basis within our field of research but are sometimes highly useful to display our spatial data in a visually appealing way - and also potentially provides...

EAGLE presentation by Gökçe Yağmur Budak

EAGLE presentation by Gökçe Yağmur Budak

On November 26, 2024, Gökçe Yağmur Budak will present her internship results on " Leveraging Data-Driven Approaches for Seismic Risk Assessment in Istanbul " at 12:30 in seminar room 3, John-Skilton-Str. 4a. From the abstract: This internship aims to create time...

social bouldering event

social bouldering event

Last Friday did we organize again a social event for all staff members and EAGLEs to get to know the new students, meet and chat with old students and just have a nice time outside the office for all staff members. We spend 4 hours together in the Rock In boulder gym...