#######################
PlantCLEF2022 datasets
#######################

Test dataset (public access on April 1, 2022):

The compressed archive containing all the images and a metadata file linking each image to its plant observation can be downloaded here:
https://lab.plantnet.org/LifeCLEF/PlantCLEF2022/test/PlantCLEF2022_test.tar.gz

Alternatively, the files can be downloaded one by one: 
 - download the metadata file first: https://lab.plantnet.org/LifeCLEF/PlantCLEF2022/test/PlantCLEF2022_test_metadata.csv
 - then find in the "image_backup_url" column the individual urls of each image with your favourite curl or wget command (for instance: cut -d';' -f4 PlantCLEF2022_test_metadata.csv | xargs wget)

The PlantCLEF2022 test dataset contains 26868 plant observations related to 55306 images, meaning that many plants are related to several images. The columns "obsid" and "image_name" reports which images are related to which plant observation. A run file to be submitted for the challenge is a csv file (with semicolon separators) and has to contain as much lines as the number of predictions, each prediction being composed of an obsid, a classid, a probability and a rank. Each line should have the following format: 
obsid;classid;probability;rank 
Due to the large number of plant observations in the test set, in order to limit the size of the run files, we ask the participants to limit the predictions to the 30 best results per observation (a maximum of 30 rows from rank 1 to 31 max for each plant observation). Thus a run file, if it contains predictions for all observations, will contain at least 26868 lines if there is only one species proposal per observation and 26868x30 lines at most if there are each time 30 species proposals.

Note that the metadata also provide a tag concerning the type of organ (leaf, flower, fruit, bark, ...) and also geographical coordinates.