| CARVIEW |
Benjamin Kellenberger
Hello, world! I am a Lecturer and permanent researcher in the People and Nature Lab at University College London (UCL), working at the intersection of Earth
observation, machine learning, and ecology to answer
questions about the distribution of species and underlying
processes.
I previously worked in the Jetz lab at
Yale
University, as well as the ECEO lab at EPFL,
Switzerland, where I researched methods to automatically
identify animals from above—using aerial imagery.
In a Nutshell
Ecologists have been contributing for decades to our understanding of where individuals of species thrive and where they don't. It turns out that most of these findings about biological nature are things we acknowledge that we don't know—perhaps more so than in any other science field (reference needed). Nonetheless, the process understanding amassed by ecologists is huge and severely underused in current methodologies.
In my research, I explore ways to integrate such latent knowledge into modelling approaches in new ways, stepping beyond hard and limiting assumptions of existing heuristics.
Digital Wildlife Conservation: What and Why
I am interested in developing and providing solutions for nature and wildlife conservation, particularly using machine learning, computer vision in conjunction with aerial or camera trap imagery. One of my major research directives to this end is the automated and interactive localisation of mammals (elephants, rhino, etc.) in aerial images. Click for more infos!
You can hover over the image to the right to see one basic example product of what machine learning models can deliver.
Wait, what's so hard about this?
Predicting bounding boxes (i.e., object detection) has long been studied in computer vision, sure. However, aerial imagery poses a completely different set of problems. If you believe this to be simple, I would like to encourage you to get a stopwatch, open up this aerial image and measure the time it takes you to find the mammals. Then, think about how hard it is—even for a computer—to do that in hundreds of thousands of such images with a high recall and low error rate...
Why even bother?
Locating animals in aerial imagery serves many purposes, such as:
- Censuses: wildlife parks, game reserves, and authorities need to periodically obtain animal counts to monitor the health of their livestock.
- Poaching prevention: even to this day, thousands of elephants, rhino, and more are poached every year in Africa. If this continues, much of our great fauna will become extinct. For example, the Black Rhino is about to die out in a few years from now...
- Knowing where the animals are and how many there are helps identifying grazing hotspots and sending out anti-poaching patrols. Plus, on a more civilised note, ecologists have great interest in these data to establish links between various ecosystem parameters, and to study the interactions between animals and flora.
- Doing all of this using aerial images enables covering larger areas from a better point of view (less occlusions). Doing it automatically means that humans need less time to annotate, the localisation quality will improve, and the costs go down.
How does this work?
As is widespread these days, much of my research uses
computer vision and machine learning (including, but not
limited to, deep learning) to identify and localise
animals. See the publications page if you
want to know more.
However, simply developing models and publishing them in
papers will not automatically change things for the
better. Instead, my work is fuelled by two critical
stances:
- Machines work best if they assist and collaborate with humans. I oppose full automation of jobs, unless the task is too dangerous or requires immense precision. Computers are good at large-scale, monotonic tasks, so let's offload this portion of the work to them, and leave the final reasoning to humans.
- I strongly believe that scientific research needs to be applied and given to the right audience, academic or not, in order to be successful. I wish to see a direct impact of my work, beyond citations.
You may want to check out my latest software projects and publications.
Other things I do
My research interests are highly interdisciplinary and
span more general vision, remote sensing and GIS topics.
In detail, I also work on domain adaptation, weakly- and
self-supervised learning, meta-learning, land cover and
land use mapping, and more.
Besides that, I have
been frequently involved in teaching (BSc and MSc
courses in GIS and machine learning for spatial data),
and I administered and maintained the computing
infrastructure of my previous research groups, including
some GPU servers back in the Netherlands.
News
I am co-advertising five PhD projects for 2025!
Projects are
distributed across two doctoral training programmes, TREES and AI-INTERVENE, and
have a wonderful team of co-supervisors from UCL, Queen Mary University
London, and EPFL Switzerland!
Enquire and apply now!
As part of the Map of Life Rapid Assessments team, I am thrilled to
announce that we have made second place in the XPRIZE Rainforest Competition this year!
This competition was five years in the making, with the aim of
advancing technology for biodiversity and ecosystem monitoring in
challenging areas like tropical rainforests.
Each of the six finalist teams from around the globe was given 24 hours to
acquire data of a predefined patch of Brazilian rainforest. The location was
kept secret until a few days before the finals in June; teams were
prohibited from setting foot into the area and had to gather data with
technology. A subsequent 48 hour window was given to analyse any data and
identify as many species as possible.
Our team employed a vast array of
technologies, from imaging and acoustic drones to air and water eDNA
collectors, and employed a human-in-the-loop scheme with taxon experts to
make sense of this data.
As head of the data science team, I was
responsible for data pre-processing, prediction (with state-of-the-art deep
learning methods), and organisation in the cloud. Our solution worked fast
and flawlessly, detecting fine-grained species in aerial images and acoustic
recordings, estimated forest functional traits in ultra high-resolution
remote sensing data, and provided innovations in all aspects, from 3D data
matching to open set species recognition.
More about our efforts at Yale.
It is my pleasure to
announce that I have officially started my new, permanent position as
Lecturer and Researcher in the People and Nature Lab at University College London (UCL)! I will be deepening my
research on data science, Machine and Deep Learning, and Earth observation,
to answer questions about our natural environment and protection thereof.
Please stay tuned if you are interested in working in these spheres and
using technology for good, as I may be hiring soon!
UCL Profile Page
Known under Map of Life Rapid
Assessments, we are one of the six finalists in the XPRIZE
Rainforest competition. Our team just completed the finals, consisting
of 24 hours of unmanned data acquisition in a square kilometre patch of
rainforest in Brazil, followed by 48 hours of data analysis, with the aim to
identify as many non-bacterial species and characterise the ecosystem as
completely as possible. Thanks to the tremendous breadth of knowledge in our
team, ranging from taxon experts and eDNA researchers to data scientists and
software engineers, we were able to set up a technology-empowered and
human-inclusive system that worked fabulously, A to Z, and gave us a wealth
of species and insights! Winners of the competition will be announced later
this year.
I will be moving labs! Starting October 1st, I will be joining the Department of Ecology and Evolutionary Biology, the Jetz Lab, at Yale University as a postdoctoral researcher! My focus will be on large-scale habitat suitability mapping with remote sensing and machine learning; so will be contributions to the Map of Life project.
New position paper in Nature Communications: Perspectives in machine learning for
wildlife conservation
We connect the dots
between ecological research on wildlife conservation and
automated heuristics from machine learning and computer
vision. We not only highlight the potential of machine
vision to assist, accelerate, enhance, and upscale
ecological research, along with success stories and
lessons learnt, but sketch an agenda to foster joint
efforts between the involved fields towards a holistic
understanding of wildlife behaviour, conservation, and
needs.
I am thrilled to have been joint first author on this landmark publication and, more importantly, honoured to have been able to work together with some of the greatest scientists and people imaginable.
TweetsA review of the use of AI/ML for wildlife conservation.
— Yann LeCun (@ylecun) February 14, 2022
"Machine and deep learning (ML; DL) bring the promise of being the right tools to scale local studies to a global understanding of the animal world."https://t.co/wM5vWVCcoF
Computer vision and machine learning for wildlife conservation is growing in importance. This is a great introduction/review. Congrats @silvia_zuffi, @TrackingActions, @sarameghanbeery, @icouzin, and many more! https://t.co/LBsashFZkn
— Michael Black (@Michael_J_Black) February 10, 2022
Development of AIDE v3.0 is on its way! Lots and lots of new features are planned. Here's just a sneak peek of some of them:
Development will take more time, but you may want to stay tuned for an exciting release!
New paper is out: 21 000 birds in 4.5 h: efficient large-scale seabird detection with machine learning!
In this work we tackled the problem of detecting and counting high-density bird colonies off West Africa. Using deep learning-based models and prior knowledge,
we were able to accurately detect 21 000 birds in drone imagery, using only 200 annotations per species and a total of 4.5 hours, from unannotated orthomosaic to prediction.
Our new paper about AIDE, the machine learning-assisted web annotation interface, is out!
It has been published in the British Ecological Society's journal Methods in Ecology and Evolution, and is openly accessible here:
Link to paper
A bit of social media presence about the work:
Twitter—YouTube (teaser video)—YouTube (extended video)
I have moved labs! From October 1 on, I will officially be an employee of the newly inaugurated Environmental Computational Science and Earth Observation Laboratory (ECEO) at EPFL in Sion, Switzerland! My duties will include all I did in Wageningen (including wildlife detection) and much more. I invite you to visit the web page (still work in progress) of our brand new lab (still in growth).
About me
I am originally from Zurich, Switzerland and also did my BSc and MSc there (University of Zurich). I completed my PhD at Wageningen University, Netherlands, with distinction "cum laude" (PhD thesis). A recording of my PhD defence from April 6, 2020, can be found here for your amusement.
I have also worked in the U.S. at Microsoft in summer 2019.
Here's a brief summary of my educational activities:
| October 2022 - July 2024 | Postdoctoral position at BGC, Yale University, United States. |
| October 2020 - September 2022 | Postdoctoral position at ECEO, École Polytechnique Fédérale de Lausanne, campus Sion, Switzerland. |
| April 7, 2020 - September 2020 | Postdoctoral position at GRS, WUR |
| May - August 2019 | Research Intern at Microsoft in the context of the AI for Earth initiative. Development of the AIDE platform. |
| Aug 2017 - April 6, 2020 | PhD candidate (cont'd) at the Laboratory of Geo-Information Science and Remote Sensing (GRS), Wageningen University (WUR), the Netherlands. |
| Feb 2016 - Jul 2017 | PhD candidate at the Remote Sensing Laboratories, University of Zurich. |
| 2015 - 2016 | Intern at the Institute of Cartography and Geoinformation, ETH Zurich. Development of the GeoVITe geodata download portal. |
| 2014 - 2015 | Intern at the Federal Office of Topography swisstopo. Development of the Swiss Map Mobile data backbone. |
| 2009 - 2014 | BSc and MSc in Geography (Remote Sensing and GIS) and Computer Science at University of Zurich, Switzerland. |
Feel free to contact me for a full CV.
People I collaborate with
Just a few of the amazing people I have or had the honour to work with (in no particular order):