ICPR 2024 Competition on EcoVision

The adoption of digital technologies has experienced a surge across various ecological domains, facilitating the collection of extensive image and video data on wildlife [REF]. This practice enhances wildlife monitoring and significantly improves efficiency in terms of time, effort, and costs. However, the collection of video and image datasets in marine environments presents distinct challenges due to the difficulties in collecting such data and the need for device retrieval for dataset download.

Ecologists employ sensing technology, known as "bio-logging" devices, on marine predator species such as penguins. These devices record videos or images while the animal is freely moving in the wild, providing valuable information about the animals' habitats (Figure 1). Through the analysis of these datasets, ecologists gain insights into how animals interact with their surroundings, with their co-specific, and with other species, as well as their responses to changing environmental conditions [REF]. This understanding is crucial for species conservation efforts and assessing the impacts of global changes on our planet.

The rapid advancement of image-based bio-logging technologies has resulted in the generation of extensive (terabytes), complex, and diverse datasets. A significant challenge lies in the considerable effort associated with manual analysis of each image. Artificial Intelligence (AI) methods and interdisciplinary research projects present solutions to address otherwise challenging tasks in ecology. This involves computer scientists developing AI approaches for these challenging datasets, to identify and correctly classify objects within the images captured by these cameras, providing ecologists with valuable insights and streamlining the analytical process. This competition provides datasets collected from wild penguins while foraging in Antarctic waters. We provide labelled image data listing objects such as the penguin carrying the bio-logging device, other penguins and/or species, type of prey these animals catch, environmental features (e.g. sea ice).

To the best of our knowledge there is no existing computational framework able to automatize the detection of these objects across the large number of images collected. The datasets generated for the “EcoVision: Unleashing AI for Marine Ecology Insights” will be made publicly available after the competition. The analytical pypelines developed during this competition will serve the scientific community for further developing inter-disciplinary projects across computer science and ecology fields and for developing frameworks for the analysis of video and image data collected from bio-logging technology across species. Below we provide details on the challenges of the competition, important dates, and short profiles of the competition organizers.