Object Counting Using a Refinement Network

To address the scale variance and uneven distribution of objects in scenarios object-counting tasks, an algorithm called Refinement Network (RefNet) is exploited. The proposed top-down scheme sequentially aggregates multiscale features, which are laterally connected with low-level information. Trained by a multiresolution density regression loss, set intermediate-density maps estimated on each feature pyramid, detailed information map gradually added through coarse-to-fine granular refinement progress to predict final map. We evaluate our RefNet three crowd-counting benchmark datasets, namely, ShanghaiTech, UCF_CC_50, UCSD, method achieves competitive performances mean absolute error root squared compared state-of-the-art approaches. further extend cell counting, illustrating its effectiveness relative counting tasks.