TASK & EVALUATION METRICS

The challenge consists of the following computer vision tasks corresponding to different plant traits.

Leaf Instance Segmentation

This segmentation task comprises of generating leaf instances automatically for plant images. The participants have to submit the zip file that contains predicted leaf instances for the test dataset. 

The metric that will be used to rank the submissions is Symmetric Best Dice (SBD). SBD is based on computation of Best Dice (BD) (Equation 1). Where, the Best Dice (BD) computes the average of maximum Dice of each leaf instances Li pr in set 1 ≤ i ≤ M with ground truth leaf instances Li gt  in set 1 ≤ j ≤ N  (Equation 2).

Leaf Counting

In this task, we ask the participants to regress leaf counts. Although, leaf count can be computed from the predicted leaf instances (previous task). The challenge in this task is to predict the leaf count directly from plant images without instance segmentation.

Participants have to submit the result according to the submission format that will be reported on the challenge website. The regression results will be evaluated with the difference in count (DiC). This metric denotes the absolute value of the difference between the number of predicted leaves and the ground truth.