The following datasets are available for your projects. Some are hosted by the research group and can be shared upon request, while others are publicly available online.
Biologging
Green turtle dataset (Martinique): Data from 14 individuals, annotated with behavioral labels.
Green turtle dataset (Martinique and French Guiana): Data from 20 individuals in Martinique and 7 individuals in French Guiana. Not annotated.
Hawksbill turtle dataset: Data from 6 individuals, annotated with behavioral labels.
Bioacoustics
Intaka acoustic dataset: Passive acoustic recordings collected at different locations and on different days over a two-week period. Total dataset size: approximately 545 GB.
An annotated subset is available for detecting Pin-tailed Whydah vocalizations (363 GB).
Bioscape PAM dataset: Passive acoustic monitoring recordings. Total dataset size: approximately 639 GB. Not annotated.
Originally downloaded from the Bioscape website.
Drone imagery
Bioacoustics
Kenya acoustic dataset: Collection of wildlife acoustic recordings from Kenya.
Greater Cape Floristic Region acoustic dataset: Sound recordings of approximately 70 species, including frogs, birds and insects.
Other wildlife datasets
LILA – Library of Images and Labels for Natural Sciences: A collection of publicly available datasets for wildlife monitoring, including camera trap images and other ecological data.
The following are examples of research questions you could investigate. You are also welcome to propose your own project, subject to approval.
Biologging
How does the amount of labelled training data affect animal behaviour classification?
Drone imagery
Can deep learning be used to estimate wildlife population size from drone images?
How does image resolution affect animal detection performance?
Bioacoustics
What is the naive occupancy of a species across different recording locations?
Does the acoustic detectability of a species vary across recording sites?
Can we estimate the relative acoustic activity of a species across different locations?
Can we investigate daily patterns in species acoustic activity?
How does the detection threshold affect species detection performance?
...
Note: Projects should be scoped to be achievable within the two-week assignment period. You do not need to use all the available data or develop a complex deep learning architecture. The focus should be on investigating a clearly defined research question through a well-designed experiment.
Credits:
Top image: This image was created with the assistance of DALL·E 3.