From Open Set to Closed Set: Supervised Spatial Divide-and-Conquer for Object Counting

Visual counting, a task that aims to estimate the number of objects from an image/video, is open-set problem by nature as population can vary in $$[0,+\infty )$$ theory. However, collected data are limited reality, which means only closed set observed. Existing methods typically model this through regression, while they prone suffer unseen scenes with counts out scope set. In fact, counting has interesting and exclusive property—spatially decomposable. A dense region always be divided until sub-region within previously observed We therefore introduce idea spatial divide-and-conquer (S-DC) transforms into problem. This implemented novel Supervised Spatial Divide-and-Conquer Network (SS-DCNet). It learn but generalize scenarios via S-DC. provide mathematical analyses controlled experiment on synthetic data, demonstrating why closed-set modeling works well. Experiments show SS-DCNet achieves state-of-the-art performance crowd vehicle plant counting. also demonstrates superior transferablity under cross-dataset setting. Code models available at: https://git.io/SS-DCNet .