Automatic pollen sensing is important to understand the local distribution of pollen in urban environments and to give personalized advice to the citizens suffering from seasonal pollen allergies to help milder the symptoms.

We present a challenging data set of labeled sequential pollen images recorded with an off-the-shelf microscope to test and improve on a variety of tasks, such as pollen detection, classification, tracking, and novelty detection. Pollen samples were gathered using a novel cyclone-based particle collector. The data set contains 16 pollen types with around 35’000 microscopic images per type and covers pollen samples from trees and grasses gathered in Graz, Austria between February and August 2020.

In addition, we share microscopic videos taken in the wild over 3 days in February and March 2020 with an automated pollen measurement system based on the same microscope technology to test and compare model performance in a natural environment. The data is available on Zenodo (https://zenodo.org/record/4120033).

N. Cao, M. Meyer, L. Thiele, O. Saukh, Pollen video library for benchmarking detection, classification, tracking and novelty detection tasks: dataset, in: DATA ’20: Proceedings of the Third Workshop on Data: Acquisition To Analysis. Association for Computing Machinery, NY (2020) 23–25