My question is about segmentation to spot light vehicles on a portion of a satellite image. I just have the satellite image and the mask (that I can open on QGIS).Although the prediction mask has some success in pinpointing vehicle locations, it suffers from two shortcomings:incomplete coverage of individual vehicles and merging detections, leading to multiple vehicles being grouped together.
What solutions would you consider to refine this mask to allow vehicle counting?