Where Are the Blobs: Counting by Localization with Point Supervision

Object counting is an important task in computer vision with applications in surveillance, traffic monitoring, and counting objects. We propose a detection-based method that outperforms regression approaches. Our contributions include: (1) a novel loss function using point-level annotations for single blob outputs per object; (2) two methods for splitting large predicted blobs; (3) state-of-the-art results on Pascal VOC and Penguins datasets, surpassing methods using depth features and bounding-box labels.