SYKE-plankton_ZooScan_2024

The SYKE-plankton_ZooScan_2024 dataset consists of over 24k expert-labeled single-specimen zooplankton images. The dataset was acquired using the ZooScan instrument applied to water samples collected from the Baltic Sea. The data is divided into 20 classes and all images were annotated by an expert taxonomist.

While the dataset can be used to train and test plankton recognition models in general, it was original composed to train and test open-set recognition (OSR) methods. We provide the splits for OSR used in the original publication. Furthermore, we provide the corresponding OSR splits for the SYKE-plankton_IFCB_2022 dataset published earlier.

If you use the SYKE-plankton_ZooScan_2024 dataset in your research, we kindly ask that you reference the following paper:

Kareinen, J., Skyttä, A., Eerola, T., Kraft, K., Lensu, L., Suikkanen, S., Lehtiniemi, M., & Kälviäinen, H. (2024). Open-Set Plankton Recognition. Out of Distribution Generalization in Computer Vision workshop at ECCV 2024.

For more details about the data collection and composition, as well as, the comparison of OSR methods on both dataset, see the paper.