Researchers at the École Polytechnique Fédérale de Lausanne (EPFL) have developed a unique video dataset named MammAlps. This dataset purely focuses on the complex behaviors of wild mammals in the Swiss Alps, applying new multi view and multi modal techniques. Inaugural Ph.D. student Valentin Gabeff heads a pioneering project with the guidance of Professors Alexander Mathis and Devis Tuia. Their aim is to expand basic understanding of wildlife dynamics in this unique alpine ecosystem.
MammAlps uses a hierarchical, controlled vocabulary to classify animal activities. It creates a hierarchical structure that separates general behaviors, like foraging and playing, from more detailed movements including walking, grooming, and sniffing. In this way, MammAlps offers a large-scale look at how different species move in correlation with their environment and one another.
Data Collection Process
The MammAlps database is based on data collected from nine camera traps, evenly distributed over three altitudinal gradients across the Swiss Alps. These traps captured more than 43 hours of unedited video over several weeks. Their work collectively produced a unique and rich visual chronicle of wildlife behavior. Ultimately, the dataset employs a multi-view, multi-human approach to studying animals in their natural habitats. With this approach, we’re able to capture a deeper understanding of what drives their behavioral habits.
The study’s research team undertook incredible and comprehensive fieldwork to collect this data. Their enthusiasm and dedication to experiencing wildlife in its natural habitat was remarkable. They were determined to document the most common species. They wanted to get footage of some more elusive species too, like alpine hares and lynx. Their team is currently working through data that has already been partially collected in 2024. To build on their recordings and study more species, they are preparing for more field studies in 2025.
Recognition and Future Work
MammAlps has received significant recognition in the scientific community, being selected as a Highlight at the IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR). Save the date! The esteemed conference will be held June 11 – 15, 2025. Through this event, Gabeff and his team will have the opportunity to present their research and ideas to an international audience. The opportunity to showcase MammAlps at such a prominent event underscores the dataset’s potential contributions to wildlife research and computer vision.
Their work doesn’t stop at explaining everyday wildlife behavior. Their broader mission is to reveal the previously unseen ways that elusive species travel and interact. By expanding their recordings and enhancing the dataset, they hope to provide valuable insights into animal behavior and ecology that could inform conservation strategies in alpine environments.
Implications for Wildlife Research
MammAlps is the new gold standard in wildlife research. Beyond the impressive scientific findings, the methodological framework it employs sets a powerful precedent for future studies focused on animal behavior. By examining these studies together, researchers can get a more holistic understanding of how animals respond to their environments and interact within them. They do this by separating the major activities from the more detailed steps.
New technologies such as camera traps allow us to study wildlife without invasive observation. This method allows them to reduce human interference with natural behaviors. This approach serves scientific inquiry very well. It furthers conservation goals by allowing research on animals in environments that best mimic their natural habitats.