Open Semester Projects and Theses

Seeing Seasons: Temporal Self-Supervised Learning from Outdoor Webcam Networks for Forest Monitoring

Photogrammetry and Remote Sensing (Prof. Schindler)

In this project you will build SSL models that learn the language of the seasons from thousands of hours of outdoor webcam footage. By exploiting the interplay between visual and meteorological signals, you will develop representations that capture how forests change over time — and use them to predict ecological events before they happen.

Keywords

Self-Supervised Learning · Temporal Representation Learning · Contrastive Learning · Predictive Architectures · Phenology · Outdoor Webcam Networks · Forest Monitoring · Environmental Computer Vision

Labels

Master Thesis

Description

Phenology is the study of recurring seasonal events in nature, such as when trees first bud in spring or shed their leaves in autumn. These transitions are sensitive indicators of climate change and have direct consequences for forest carbon cycling, biodiversity, and ecosystem management. Operational forest management demands not just observation but early prediction of such events, enabling real-time stress monitoring and timely interventions. Networks of phenocams — high-resolution near-surface cameras capturing tree crowns and canopies every 30 minutes year-round — provide a rich source of such data, with 34 sites across Switzerland and over 700 sites globally. Yet the vast majority of this imagery remains unlabeled, leaving it underexploited beyond simple color-based indices. This thesis explores self-supervised learning (SSL) on these spatio-temporal image sequences, leveraging the interplay between visual and meteorological signals to learn representations that faithfully reflect the phenological trajectory of trees across sites, species, and years — ultimately enabling more accurate and timely phenological monitoring at scale. The central research question of this thesis is: what temporal self-supervised learning objectives can be designed to learn representations that improve phenological monitoring? The intersection of temporal self-supervision and weather as an auxiliary learning signal is a relatively underexplored space, with potential extensions to satellite remote sensing time series. Two promising directions are suggested as potential starting points, though the student is encouraged to engage with the literature and help shaping the methodological choices. One direction would consist in a climate-informed contrastive approach, using meteorological state (e.g., accumulated temperature, day length, recent precipitation) as a proxy for phenological stage to construct training pairs across sites and years. Alternatively, JEPA-style predictive architectures training the model to anticipate future canopy states from current context would directly align pre-training with the early prediction task.

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-30 , Earliest start: 2026-06-01 , Latest end: 2027-06-01

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Schindler Konrad

Topics Information, Computing and Communication Sciences , Biology

High-Resolution Air Pollution Mapping: Downscaling Sentinel-5P Data using Multi-Modal Deep Learning

Photogrammetry and Remote Sensing (Prof. Schindler)

This project investigates the use of advanced deep learning techniques (e.g., diffusion models or vision transformers) to perform spatial super-resolution of satellite-derived air quality data. By fusing coarse Sentinel-5P NO2 observations with high-resolution auxiliary data (such as optical imagery, land cover, elevation, and traffic network data), the project will develop a robust model capable of estimating urban air pollution at the neighborhood level.

Keywords

Deep Learning, Remote Sensing, Air Quality, Multi-Modal, Super-Resolution, Computer Vision.

Labels

Semester Project , Master Thesis

Description

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-25 , Earliest start: 2026-06-01 , Latest end: 2027-04-30

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Scheibenreif Linus

Topics Information, Computing and Communication Sciences , Earth Sciences

Implicit Earth: Neural Compression of Geospatial Foundation Embeddings

Photogrammetry and Remote Sensing (Prof. Schindler)

Global geospatial embedding products provide powerful, pre-computed latent representations of the Earth, but their discrete, grid-based nature requires significant storage and limits sub-pixel analysis. This thesis proposes training an Implicit Neural Representation (INR) to "memorize" these pre-computed embeddings. By training a coordinate-based network to map spatial coordinates to high-dimensional semantic vectors, this project investigates whether the entire latent representation of a region, country, or continent can be compressed into the network's weights while preserving the high-frequency semantics necessary for downstream tasks.

Keywords

Geospatial embeddings, implicit neural representation, location encoder, compression, foundation models.

Labels

Semester Project , Master Thesis

Description

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-25 , Earliest start: 2026-06-01 , Latest end: 2027-04-30

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Scheibenreif Linus

Topics Information, Computing and Communication Sciences , Engineering and Technology

Dense spatiotemporal reconstruction of river water levels with deep learning

Photogrammetry and Remote Sensing (Prof. Schindler)

River water level estimates from satellites are important for hydrological modelling, flood forecasting, and freshwater monitoring, but measurements from in situ gauges as well as observations from satellite altimeters remain sparse in space and/or time. This project applies neural networks trained directly on multi-source observations to reconstruct dense, continuous water level records for major river systems worldwide.

Keywords

Deep learning, Transformer neural networks satellite altimetry, hydrology, time series, graph neural networks

Labels

Semester Project , Collaboration , Master Thesis

Description

Rivers are critical and dynamic components of the Earth's water cycle, yet their comprehensive monitoring as well as predicting their dynamics remains challenging. As opposed to river discharge, water surface elevation (WSE) can be directly observed at a large scale from space by altimeters. In comparison, in-situ flow gauges maintained by national agencies (e.g., ANA in Brazil) provide temporally dense but geographically extremely sparse records. Satellite altimetry offers global yet sparse coverage in ungauged areas. The recent SWOT (Surface Water and Ocean Topography) mission introduces unprecedented spatial density through its wide-swath InSAR sensor, but its record only starts in 2023 and it is scheduled for decommissioning after 3.5 years of operation. Synthesizing these heterogeneous, non-overlapping data sources into a homogeneous dense product is an open problem in hydrology that could greatly improve hydrological modelling and impact downstream applications. This project develops neural networks trained via a masked reconstruction objective to jointly assimilate SWOT and classical altimetry observations in the form of timeseries, as well as in-situ gauge data, and produce dense daily WSE estimates for all river reaches in major river basins. The model exploits the tree-shaped topology of river networks using the SWORD river database to learn physically consistent spatiotemporal dynamics. Current work focuses on the Amazon basin and targets a reconstruction spanning 2016–2025, with planned extension to other major rivers globally. **Methods** The project uses transformers and LSTMs trained with masked reconstruction, on timeseries from multi-source observational data from SWOT, various classical altimetry satellites, and in-situ gauges (e.g. from ANA). River topology (tree-shaped) from the SWORD database is used to build samples and serves as a condition to the model. The method is implemented with PyTorch and PyTorch Lightning and is trained on HPC infrastructure. A working data processing pipeline, model implementation, and preliminary results for the Amazon basin are already in place. **Wanted** Highly motivated student with a strong background in computer science or machine learning, interested in scientific research at the intersection of deep learning and Earth system science. Experience with PyTorch is expected. Interest in geophysical or environmental applications, hydrology or Earth science is a plus. You will get to - Gain hands-on experience with neural networks, applied to real-world observational datasets from frontier satellite missions. - Work at the forefront of data-driven Earth system science with direct applications to climate monitoring and hydrology. - Expand your network by collaborating with an interdisciplinary team of researchers from ESA Φ-lab and ESA Earth System Science Hub. - Be a co-author on publications resulting from this work. The project includes a **research visit** at ESA Φ-lab, ESRIN, Frascati, Italy. The visit should ideally take place in July--September 2026. We provide a reimbursement for travel costs and accommodation. We look forward to receiving your application with the following documents: (1) your CV and (2) a max. 300-word motivation for why you are interested.

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-23 , Earliest start: 2026-07-01 , Latest end: 2027-01-31

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Schindler Konrad

Topics Information, Computing and Communication Sciences , Engineering and Technology , Earth Sciences

Synthesizing Aerial Images from Building Roof Geometry

Photogrammetry and Remote Sensing (Prof. Schindler)

This thesis explores a generative pipeline for creating accurate synthetic aerial imagery using existing geospatial data such as cadaster maps and OpenStreetMap (OSM). The proposed method investigates conditioning-based generative approaches to adapt building roof structures so they align with available labels while preserving the original aerial image style, including roof colors, textures, lighting and surroundings. The motivation is that many existing datasets contain misalignments and outdated annotations when compared to real imagery. By generating corrected yet realistic training data, this work aims to support future tasks such as roof segmentation, vectorization, and automated roof modeling.

Keywords

Synthetic Data Generation Remote Sensing Diffusion Models Geospatial Data Deep Learning Urban Mapping

Labels

Semester Project , Master Thesis

Description

Obtaining high-quality training data for Deep Learning models in remote sensing remains a significant challenge. Existing datasets, including cadaster data, OSM, and official government records, often exhibit discrepancies when compared to real-world aerial or satellite imagery. These misalignments arise due to differences in data acquisition methods, temporal inconsistencies, and inaccuracies in manually or semi-automatically generated vector datasets. This thesis investigates different conditioning strategies for diffusion-based generative models using structured inputs such as roof masks, polygons, or vector representations derived from geospatial data. In particular, spatial conditioning methods such as ControlNet [1], parameter- efficient adaptation techniques such as LoRA [2], and prompt-based conditioning approaches are considered as potential solutions. The focus is on maintaining high visual realism while enforcing strong geometric consistency between generated imagery and input roof structures. The resulting synthetic data is intended to support the development of more accurate methods for roof segmentation, roof vectorization, and automated roof modeling, while reducing the need for costly manual data correction. [1] Zhang, L., Rao, A. & Agrawala, M. (2023). Adding conditional control to text-to-image diffusion models. ICCV. [2] Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., ... & Chen, W. (2022). Lora: Low- rank adaptation of large language models. ICLR. [3] Rombach, R., Blattmann, A., Lorenz, D., Esser, P. & Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. CVPR. **Requirements** - Background in data science and machine learning; - Knowledge of deep machine learning; - Programming in Python/PyTorch; - Experience with contemporary network architectures (vision transformers, diffusion models) is a plus **Location** On-site at the Photogrammetry and Remote Sensing Laboratory (Prof. Schindler, ETH Zürich)

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-23 , Earliest start: 2026-09-01 , Latest end: 2027-08-31

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Schindler Konrad , Nicolicioiu Armand

Topics Information, Computing and Communication Sciences , Engineering and Technology

Geospatial AI: Designing Embedding Models for Multimodal Satellite Time Series

Photogrammetry and Remote Sensing (Prof. Schindler)

In this project, you will develop and evaluate new AI architectures for learning compact embeddings from multimodal satellite time series. The goal is to create models that encode long sequences of Earth observation data into efficient, shareable representations that retain all relevant information for downstream analysis. The project is executed as an onsite project at IBM Research Zurich in collaboration with ETH and provides access to HPC infrastructure.

Keywords

Earth sciences, embedding models, multimodal data, time series, representation learning, computer vision

Labels

Semester Project , Master Thesis

Description

Goal

Contact Details

More information

Open this project...

Published since: 2026-04-22 , Earliest start: 2026-08-01 , Latest end: 2027-03-31

Applications limited to ETH Zurich , University of Zurich

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Blumenstiel Benedikt

Topics Information, Computing and Communication Sciences

Reliable Point Tracking in Presence of Non-Lambertian Surfaces

Photogrammetry and Remote Sensing (Prof. Schindler)

In collaboration with the University of Bologna, you will develop and evaluate novel point trackers robust to the presence of non-Lambertian objects.

Keywords

Point Tracking, Non-Lambertian Reflection, Specular and Transparent Objects, Multi-layered Scene Understanding, Video 3D Understanding

Labels

Master Thesis

Description

Point tracking is a specific task in computer vision that involves identifying and following the trajectory of a set of individual pixels, or a set of "interest points" or, in some cases, the totality of the pixels [1] across a sequence of video frames [2]. This process typically relies on the assumption of Lambertian reflectance, where the appearance of an object remains constant regardless of the viewing angle. However, real-world scenarios frequently involve non-Lambertian surfaces, such as metallic, glassy, or polished materials. These surfaces exhibit specular highlights, reflections, and view-dependent color shifts that violate the brightness constancy constraint. This constitutes a serious challenge for conventional trackers, which often fail on these objects because they interpret moving reflections as changes in the object’s geometry or position, leading to significant "drift" or total loss of the target points. A particular setting is represented by multi-layered scenes [3], characterized by the presence of multiple transparent surfaces. The goal of tracking itself, in this case, becomes ambiguous – given a glass door, should the tracker follow the trajectory of the points behind it, should it properly track the points on the door’s surface, or both? This project aims to study the possibilities enabled by this peculiar setting, and to develop a robust foundation model for point tracking that inherently accounts for non-Lambertian properties and allows to track points differently when selecting one among multiple layers in the scene. [1] https://arxiv.org/abs/2506.07310, Harley et al., “AllTracker: Efficient Dense Point Tracking at High Resolution”, ICCV 2025; [2] https://arxiv.org/abs/2410.11831 Karaev et al., “CoTracker3: Simpler and Better Point Tracking by Pseudo-Labelling Real Videos”, ICCV 2025; [3] https://arxiv.org/abs/2409.05688, Wen et al., “LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow”, ECCV 2024; **Requirements**: • Experience with Python, PyTorch, Vision Transformers, and the Hugging Face ecosystem; • Deep understanding of the references mentioned above and other related literature; • Examples of code written for other projects hosted on GitHub; • Problem-solving mindset: the ability to explore technical literature, cross-reference relevant information, and synthesis of novel ML architectures • Ways to stand out: hands-on experience with HPC systems and job scheduling, good knowledge of Computer Vision and Multi-View Geometry

Goal

Contact Details

More information

Open this project...

Published since: 2026-03-13 , Earliest start: 2026-04-01 , Latest end: 2026-11-30

Applications limited to ETH Zurich , EPFL - Ecole Polytechnique Fédérale de Lausanne , University of Zurich

Organization Photogrammetry and Remote Sensing (Prof. Schindler)

Hosts Sakaridis Christos

Topics Information, Computing and Communication Sciences , Engineering and Technology

Tree species identification using deep learning

Forest Resources Management

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

Open this project...

Published since: 2026-02-13 , 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