Conceptual

Hierarchical Panoptic Segmentation of Crops and Leaves with Boundary Loss

A computer-vision method that adapts the Mask2Former panoptic-segmentation architecture with a second transformer decoder so it segments whole plants and their individual leaves at once, while adding focal loss and boundary loss to better capture the tiny, class-imbalanced regions of leaves and weeds. Students learn how boundary-based loss integrals over region interfaces and focal down-weighting of easy pixels improve segmentation and leaf-counting accuracy in precision agriculture without increasing inference cost.