NEWS & EVENTS

PlantCLEF is one of five challenges within the LifeCLEF 2025 initiative, and it stands out as a project funded by GUARDEN  and MAMBO projects. This funding underscores the challenge’s strategic role in advancing AI-driven plant identification and ecosystem monitoring.

LifeCLEF itself is an international research initiative hosted within the CLEF evaluation forum. It fosters cutting-edge AI applications to understand and catalog biodiversity through a series of annual competitions focused on animals, birds, fungi, plants, and geospatial species distribution.

Within this landscape, PlantCLEF 2025 is dedicated to multi-species plant identification in high-resolution vegetation plot images. Participants are invited to tackle a multi-label classification task: given a 0.5 m² quadrat image featuring mixed flora, the goal is to predict every plant species present. The key challenge lies in bridging the gap between the training data composed of single-plant images and the test data, which consists of dense, multi-species vegetation photos.

Training data comes from two main sources. First, a subset of Pl@ntNet imagery covers around 7,800 plant species from southwestern Europe, totalling approximately 1.4 million labelled images. This is further supplemented with curated data from GBIF to bolster species with fewer examples. Second, a large secondary dataset of several hundred thousand unlabelled vegetation cover images from LUCAS is provided to support self-supervised learning approaches. The test set, on the other hand, comprises expert-curated quadrat images representing diverse Pyrenean and Mediterranean contexts.

The PlantCLEF 2025 registration opened in December 2024, the competition kicked off on March 1 of 2025, and the submission deadline was May 19, 2025. Working note papers were then due by June 7 and camera-ready versions finalized by June 30.  The results and participant findings will be presented during CLEF 2025, scheduled for September 9–12 in Madrid!

PlantCLEF also encourages innovative learning paradigms—self-supervised, semi-supervised, and unsupervised methods are strongly supported, and pretrained models are made available to participants. The rules allow for the inclusion of external datasets or metadata, provided that participants also submit runs using only the supplied data.

LifeCLEF’s broader mission is to drive innovation at the intersection of AI and biodiversity through rigorous, data-driven challenges. PlantCLEF 2025 exemplifies this goal by focusing on one of today’s most complex tasks in ecological monitoring which is to accurately identify multiple plant species within a single image. The support from GUARDEN and MAMBO is then evident, considering their objectives of developing scalable, automated biodiversity assessment tools to guide conservation and environmental decision-making.

As the PlantCLEF challenge transitions into publication and community discussion, its contributions promise to impact not only academic research but also real-world ecological survey practices. The collaborative outputs, including working notes, shared datasets, and open-source models, will continue to reinforce how AI can amplify environmental stewardship and help safeguard plant diversity worldwide.