Start Date: March 21 End Date:  24 May Competition URL: HuggingFace

Motivation

Vegetation plot inventories are essential for ecological studies, enabling standardized sampling, biodiversity assessment, long-term monitoring and remote, large-scale surveys. They provide valuable data on ecosystems, biodiversity conservation, and evidence-based environmental decision-making. Plot images are typically 0.5x0.5 meter size, and botanists meticulously identify all the species found there. In addition, they quantify species abundance using indicators such as biomass, qualification factors, and areas occupied in photographs. The integration of AI could significantly improve specialists' efficiency, helping them extend the scope and coverage of ecological studies.

Task Description

The task will be evaluated as a multi-label classification task that aims to predict all the plant species visible on the high-resolution plot images. The main difficulty of the task lies in the shift between the test data (high-resolution multi-label images of vegetation plots) and the training data (single-label images of individual plants).

As an example, the following pictures depict a typical vegetative plot where a botanist has recorded 8 species (Cardamine resedifolia L., Festuca airoides Lam., Pilosella breviscapa (DC.) Soják, Lotus alpinus (Ser.) Schleich. ex Ramond, Poa alpina L., Saxifraga moschata Wulfen, Scorzoneroides pyrenaica (Gouan) Holub, Thymus nervosus J.Gay ex Willk). Today, with the rise of biodiversity informatics, collaborative platforms such as inaturalist, plantnet and gbif, potentially many photographs of individual plants on these species have been shared. However, there is very little vegetative plot data, as this is a too tedious task. The idea of the challenge is to see to what extent these individual plant data can nevertheless help to build/train multi-species systems/models.