Pollen Grain Classification Challenge (ICPR 2020) Call for Papers

Pollen Grain Classification Challenge (ICPR 2020)

Click here to modify your subscriptions or to unsuscribe from future emails.

Challenge website: https://iplab.dmi.unict.it/pollenclassificationchallenge/

Main contact: Alessandro Ortis ortis@dmi.unict.it

Brief description/Scope

The aim of the proposed challenge is the automatic classification of
pollen grain images exploiting the largest dataset of microscope
pollen grain images, collected from aerobiological samples. The
microscope images of the samples have been digitalized and processed
through a proper image processing pipeline to detect and extract four
classes of objects, including three species of pollen grain and an
additional class of objects that could be often mis-classified as
pollen (e.g., air bubbles, dust, etc.).

More than 13.000 objects have been detected and labelled by
aerobiology experts.

Task description and material

A set of images related to objects detected in microscope images will
be given to participants, where each depicted object belongs to one of
the defined classes. Segmentation mask of each object will also be
provided, in order to allow methods that exploit the localization of
the salient regions in images (e.g., attention). In addition, the
output of the segmentation will also be provided, for visualization
purposes.

Participants are requested to classify correctly the highest number of
objects. Moreover, the proposed methods should also address with the
imbalance in the data, which represents one challenge of pollen grain
classification in real-world scenarios.

Test images will be released, without the ground truth classes, nor
the segmentation masks. Participants are requested to upload the
results of classification according to the submission format that will
be reported on the challenge website.

The competition website will provide additional information about the
data, material and code examples that perform a preliminary data
exploration.

Test classification results will be evaluated using the weighted F1
score for quantitative evaluation. This metric has been selected
considering the imbalance in the data. This function calculates the F1
metrics for each class, and their average weighted by support (the
number of true instances for each class). This alters the unweighted
F1 score to account for label imbalance.

Participants will present their works during a competition track at
ICPR 2020, co-located with the conference, where competition results
will be presented and discussed. In addition, the authors of the best
selected works will be invited to contribute to a joint paper on the
addressed challenge. The paper will include the most significative
works and their findings, will be submitted to a valuable Journal.

Important Dates

    Registration opening: April 10
    Training data available: April 20
    Testing data available: May 18
    Test result submission deadline: May 22
    Announcement of the evaluation results: June 30

Contacts

For any further information, please contact us at ortis@dmi.unict.it
and go to the challenge website
https://iplab.dmi.unict.it/pollenclassificationchallenge/.

Team that will run the Challenge

Image Processing experts
     Sebastiano Battiato, Full Professor, Università degli Studi di Catania, Italy
     Alessandro Ortis, Postdoctoral Researcher, Università degli Studi di Catania, Italy
     Francesco Guarnera, PhD Student, Università degli Studi di Catania, Italy
      Francesca Trenta, PhD Student, Università degli Studi di Catania, Italy

Aerobiology experts
      Consolata Siniscalco, Full Professor, University of Turin, Italy
      Lorenzo Ascari, PhD Student, University of Turin, Italy

Agronomy experts
      Eloy Suárez | Agri Competence Center | Ferrero HCo
      Tommaso De Gregorio | Agri Competence Center | Ferrero HCo

 

Acknowledgement

The research has been carried out thanks to the collaboration with
Ferrero HCo, that financed the project and allowed the collection of
aerobiological samples from hazelnut plantations.