Open Semester Projects and Thesis
Note: Topics labeled as "semester project" are suitable for the "Geomatics Project I" or "Geomatics Project II" in the MSc curriculum 2022.
Remote Sensing Image Instance Segmentation
This project aims to develop a robust instance segmentation model for remote sensing imagery. The goal is to adapt existing architectures (like Mask2Former) and leverage foundation models (like SAM) to accurately detect and segment individual objects (e.g., buildings, vehicles) in large-scale aerial datasets like iSAID.
Keywords
instance segmentation, remote sensing, aerial imagery, deep learning, computer vision, foundation models
Labels
Semester Project
Description
Goal
Contact Details
More information
Published since: 2025-10-30 , Earliest start: 2025-10-31 , Latest end: 2027-09-30
Applications limited to ETH Zurich
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Chen Yuxing
Topics Engineering and Technology
Distilling a Semantic Change Representation from Mono-Temporal Pretrained Features for Unsupervised Change Analysis
This project develops an unsupervised framework to learn a dense semantic representation of change from bi-temporal remote sensing images. By designing a "Change Distillation" module that processes features from a pretrained backbone (e.g., DINOv2), the goal is to generate a single "change vector" per pixel that explicitly encodes the "from-to" transaction, enabling the automatic discovery of different semantic change types.
Keywords
unsupervised change detection, semantic change analysis, remote sensing, feature distillation, pretrained models, representation learning
Labels
Semester Project , Bachelor Thesis
Description
Goal
Contact Details
More information
Published since: 2025-10-30 , Earliest start: 2025-10-31 , Latest end: 2027-09-30
Applications limited to ETH Zurich
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Chen Yuxing
Topics Engineering and Technology
Benchmarking hyperspectral remote sensing
Hyperspectral sensors capture the electromagnetic spectrum in fine detail, enabling the mapping of specific physical phenomena on Earth. With the recent start of the EnMAP mission, large amounts of hyperspectral remote sensing data are becoming publicly available. In this project, we would like to evaluate the benefits of hyperspectral data over more commonly used and widely available multi-spectral remote sensing imagery.
Keywords
Hyperspectral, multispectral, benchmarking, deep learning, Earth observation
Labels
Bachelor Thesis
Description
Goal
Contact Details
More information
Published since: 2025-10-30 , Earliest start: 2025-12-01 , Latest end: 2026-12-01
Applications limited to ETH Zurich
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Scheibenreif Linus
Topics Information, Computing and Communication Sciences , Engineering and Technology
Unsupervised Remote Sensing Image Segmentation using Slot Attention and Pretrained Features
This project develops an unsupervised segmentation framework for remote sensing (RS) imagery. By adapting the Slot Attention mechanism and integrating features from large pretrained models (like DINOv2 or Prithvi), the goal is to automatically discover and segment detailed land cover classes without pixel-level labels.
Keywords
unsupervised semantic segmentation, remote sensing, slot attention, self-supervised learning, feature extraction
Labels
Semester Project , Bachelor Thesis
Description
Goal
Contact Details
More information
Published since: 2025-10-30 , Earliest start: 2025-10-31 , Latest end: 2027-09-30
Applications limited to ETH Zurich , Department of Civil, Environmental and Geomatic Engineering
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Chen Yuxing
Topics Engineering and Technology
Structured Roof Geometry Generation from Remote Sensing Imagery Using Conditional Diffusion Models
This thesis explores a generative pipeline for structured roof geometry reconstruction from remote sensing imagery (RGB and/or DSM). Using conditional diffusion models, we estimate roof primitives and their geometric relationships, representing them as a graph with raster-based geometry, relational edges, and vectorized outlines.
Keywords
City modeling, Machine Learning, Layout Graphs, Remote Sensing
Labels
Semester Project , Master Thesis
Description
Goal
Contact Details
More information
Published since: 2025-09-29 , Earliest start: 2025-05-04 , Latest end: 2026-04-30
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Schindler Konrad
Topics Information, Computing and Communication Sciences , Engineering and Technology
Better maps of plant functional traits - towards planttraits.earth v2
Functional traits describe biophysically relevant properties of plants and form an important basis for understanding ecosystem dynamics and the Earth system. Planttraits.earth has recently produced global high-resolution maps of many plant traits (some of which have never before been mapped globally), by combining field data from plant scientists, crowd-sourced data from citizen scientists, and remote sensing imagery. The present project will develop methods to improve those maps and bring plant trait mapping to the next level.
Keywords
Vegetation mapping, plant functional traits, deep machine learning, Earth observation
Labels
Semester Project , Master Thesis
Description
Goal
Contact Details
More information
Published since: 2025-09-29 , Earliest start: 2025-05-04 , Latest end: 2026-04-30
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Scheibenreif Linus
Topics Information, Computing and Communication Sciences , Engineering and Technology , Biology
Tree species identification using deep learning
Tree species maps are crucial for effective forest management, biomass assessment, and biodiversity monitoring. Remote sensing products offer flexible and cost-effective ways to assess forest characteristics, while deep learning methods promise high predictive accuracy and transformative applications in forestry. This study aims to apply novel deep learning approaches to detect and identify individual trees and tree species in mixed forests. By addressing the challenges of tree species identification, this research will enhance biodiversity assessment, forest resilience understanding, and management strategies.
Keywords
Tree species identification, computer vision, CNN
Labels
Semester Project , Master Thesis , ETH Zurich (ETHZ)
Description
Goal
Contact Details
More information
Published since: 2025-09-15 , Earliest start: 2025-01-06 , Latest end: 2026-08-31
Applications limited to ETH Zurich , Department of Environmental Systems Science , Department of Civil, Environmental and Geomatic Engineering , Institute of Geodesy and Photogrammetry
Organization Forest Resources Management
Hosts Schindler Konrad , Beloiu Schwenke Mirela , Hangartner Ariane
Topics Agricultural, Veterinary and Environmental Sciences , Information, Computing and Communication Sciences , Engineering and Technology
Distillation of Remote Sensing Foundation Models
Foundation models trained on broad geospatial data are increasingly used for the interpretation of remote sensing images. Currently available foundation models typically target specific remote sensing modalities and are only used after fine-tuning on a labeled dataset for the task of interest. For large foundation models, fine-tuning and subsequent inference incur high computational cost. This can potentially be alleviated by model distillation techniques that aim to compress the pre-trained model into a smaller version. This project aims to apply model distillation methods to remote sensing foundation models with the aim of generating small and powerful models suitable for different remote sensing applications.
Keywords
remote sensing foundation models, model distillation, deep machine learning, Earth observation
Labels
Semester Project , Bachelor Thesis , Master Thesis
Description
Goal
Contact Details
More information
Published since: 2025-08-17 , Earliest start: 2025-05-01 , Latest end: 2026-04-30
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Schindler Konrad
Topics Information, Computing and Communication Sciences , Engineering and Technology
Clear And Look: Diffusion-Based 3D Estimation and Defogging
Estimating 3D scene geometry in adverse weather, particularly fog, is critical for outdoor applications such as autonomous driving. In foggy environments, robust depth perception enables safe navigation and planning despite low visibility. This thesis explores a foundation model approach using latent diffusion models for joint 3D estimation and defogging. It will leverage the physics of foggy image formation via a forward model and incorporate it into a pretrained latent diffusion framework. The goal is to produce both accurate 3D outputs and photorealistic defogged images from foggy inputs.
Keywords
Latent Diffusion Models, 3D Scene Geometry, Adverse Weather, Forward Modeling, Defogging, Autonomous Driving
Labels
Master Thesis
Description
Goal
Contact Details
More information
Published since: 2025-08-07 , Earliest start: 2025-10-01 , Latest end: 2025-12-18
Organization Photogrammetry and Remote Sensing (Prof. Schindler)
Hosts Narnhofer Dominik , Sakaridis Christos
Topics Information, Computing and Communication Sciences , Engineering and Technology