V. Zermatten

Valérie Zermatten

PhD Candidate in Earth Observation and Machine Learning · ECEO, EPFL · Sion, Switzerland

I'm looking for a postdoctoral position abroad where I could use my AI skills with Remote Sensing to investigate biodiversity.

I'm a PhD Candidate working at the intersection of Remote Sensing, Machine Learning and Ecology. I combine both computer vision and natural language processing to integrate multiple data sources (e.g., species observations, satellite images, text or and environmental rasters). My main focus during my PhD is on integrating knowledge from text to better understand remote sensing images. Specifically, I introduced knowledge from species habitat descriptions in Wikipedia to better map their ecosystems and various related concepts. In general, my goal is to use recent breakthroughs in AI to try to address the environmental challenges our planete is facing.

I'm supervised by Prof. Devis Tuia, head of the Environmental Computational Science and Earth Observation Laboratory (ECEO) at EPFL, Sion (Switzerland), and Prof. Diego Marcos , Junior Professor in the Evergreen team at INRIA, Montpellier (France). I also collaborate with the Federal Office of Topography swisstopo, to develop interactive methods for mapping Switzerland at scale. Previously, I obtained my Bachelor and Master degree at EPFL in the faculty of Environmental Science and Engineering. I spent the 3rd year of my Bachelor at Nanyang Technological University (NTU) in Singapore.

News

2026

May 2026 Expected PhD defense

2025

Dec 11Presentation and panel discussion at the online event AI for Conservation organized by WILDLABS.

Sep 26 Panelist at the Swiss {AI} Weeks online event AI for Space Lifting Off with AI From Earth to Space.

Aug 25 Presentation of EcoWikiRS at the Swiss Remote Sensing Days in Dübendorf, Switzerland.

Jun 26 Presentation of EcoWikiRS at the European Space Agency Living Planet Symposium in Vienna, Austria.

Jun 11 Lightning talk and poster for EcoWikiRS at the CVPR EarthVision Workshop in Nashville, USA.

May 21 Online presentation of EcoWikiRS at the Wiki Workshop 2025.

Mar 13 Presentation of EcoWikiRS at the AI Imagery Workshop in Le Grand Bornand, France.

Feb 27 Presentation of TACOSS at Maison de la Télédétection in Montpellier, France.

2024

Sep 26 Presentation at INRIA in Montpellier, France.

Jul 7 Tutorial at the International Geoscience and Remote Sensing Symposium in Athens, Greece.

Apr 18 Presentation at the European Geoscience Union on Large-scale mapping of environmental variables by combining ground observations, remote sensing, and machine learning in Vienna, Austria.

Mar 25 Presentation and panel discussion at Applied Machine Learning Days in the AI from the Sky session in Lausanne, Switzerland.

Mar 22 Presentation at the swisstopo Kolloquium Geodaten und KI in Wabern, Switzerland.

2023

Jul 21 Presentation of TACOSS at the International Geoscience and Remote Sensing Symposium in Los Angeles, USA.

Research

Coming soon.

Under review.

V. Zermatten, D. Marcos, D. Tuia · under review as of Jan 2026

We systematically evaluate how text can contribute to the prediction of environmental variables, such as climatic, edaphic, geologic or land use properties, when combined with remote sensing imagery. Details soon.

EcoWikiRS: Learning Ecological Representations of Satellite Images from Weak Supervision with Species Observations and Wikipedia

V. Zermatten, J. Castillo-Navarro, P. Jain, D. Tuia, D. Marcos · CVPR EarthVision Workshop, 2025

We propose a method to learn ecological properties of aerial images by learning an alignment with species habitat descriptions from Wikipedia. Such supervision improves the recognition of ecosytems, and can be used to map natural concepts at scale.

Project Page & Repo PDF Dataset

EcoWikiRS in 5 min:

Big thanks to Prof. Grace W. Lindsay from New York University for making the video.

Learning transferable land cover semantics for open vocabulary interactions with remote sensing images

V. Zermatten, J. Castillo-Navarro, D. Marcos, D. Tuia · ISPRS Journal of Photogrammetry and Remote Sensing, 2025

We introduce TACOSS, a model for land cover mapping from text: TACOSS leverages the common sense knowledge captured by language models and is capable mapping any land cover based on a textual description, beyond the set of land cover labels seen at training time.

Project Page & Repo PDF

Land Cover Mapping From Multiple Complementary Experts Under Heavy Class Imbalance

V. Zermatten, X. Lu, J. Castillo-Navarro, T. Kellenberger, D. Tuia · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024

We introduce a Multiple Complementary Experts (MCE) model to improve predictions of rare classes in semantic segmentation. We apply this approach to land cover mapping in alpine regions.

PDF

Prompt–RSVQA: Prompting visual context to a language model for Remote Sensing Visual Question Answering

C. Chappuis, V. Zermatten, S. Lobry, B. Le Saux, D. Tuia · CVPR EarthVision Workshop, 2022

We introduce a new approach for remote sensing visual question answering (RS-VQA), i.e. given a remote sensing image and a question about it, the VQA task consists in predicting a text answer. We propose to convert the visual information to text, i.e. a prompt, before injecting it into a language model to obtain an answer.

PDF

Hobbies

When I step away from my computer, I enjoy outdoor activities in the nearby mountains: running, hiking, camping, skiing and ski touring, and mushroom or blueberry picking...

Indoors, my passion for plants has taken over my living room🌿. I kept a 350 L paludarium for many years and recently transformed it into a terrarium, where I grow colorful plants . I used to keep and breed killifish 🐟. I also enjoy cooking, watercolor painting, scuba diving, and playing volleyball, which I have been doing for over 10 years.