Mixture of Counting CNNs: Adaptive Integration of CNNs Specialized to Specific Appearance for Crowd Counting

• DOI: 10.48550/arxiv.1703.09393
• Publication Date: 2017-01-01
• Authors: Shohei Kumagai; Kazuhiro Hotta; Takio Kurita
• Keywords: Robustness; Cell counting

This paper proposes a crowd counting method. Crowd is difficult because of large appearance changes target which caused by density and scale changes. Conventional methods generally utilize one predictor (e,g., regression multi-class classifier). However, such only can not count targets with well. In this paper, we propose to predict the number using multiple CNNs specialized specific appearance, those are adaptively selected according test image. By integrating CNNs, proposed method has robustness experiments, confirm that lower error than CNN integration fixed weights. Moreover, each automatically appearance.