Learn from Object Counting: Crowd Counting with Meta‐learning

The objective of crowd counting is to learn a counter that can estimate the number people in single image. So far, most proposed work evaluates density by fitting constructed map corresponding sample. performance those algorithms depends on large amount carefully prepared data. However, significant problem with data sets difficulty labeling. To address such situation, utilizing object few-shot scenes considered and an efficient algorithm extract meta-information proposed, thus improving accuracy convergence rate tasks. Specifically, network trained only tasks different domains during meta-training phase. Then, meta-counter testing meta-testing stage. Experimentally, it demonstrated above way improves converge three datasets when ten-type