Spatiotemporal Dilated Convolution With Uncertain Matching for Video-Based Crowd Estimation

In this paper, we propose a novel SpatioTemporal convolutional Dense Network (STDNet) to address the video-based crowd counting problem, which contains the decomposition of 3D convolution and the 3D spatiotemporal dilated dense convolution to alleviate the rapid growth of the model size caused by the Conv3D layer. Moreover, since the dilated convolution extracts the multiscale features, we combine the dilated convolution with the channel attention block to enhance the feature representations. Due to the error that occurs from the difficulty of labeling crowds, especially for videos, imprecise or standard-inconsistent labels may lead to poor convergence for the model. To address this issue, we further propose a new patch-wise regression loss (PRL) to improve the original pixel-wise loss. Experimental results on three video-based benchmarks, i.e., the UCSD, Mall and WorldExpo'10 datasets, show that STDNet outperforms both image- and video-based state-of-the-art methods. The source codes are released at https://github.com/STDNet/STDNet.

Automated summary

This paper proposes a novel SpatioTemporal convolutional Dense Network (STDNet) for video-based crowd counting problem. The network decomposes 3D convolution and utilizes 3D spatiotemporal dilated dense convolution to alleviate the rapid growth of model size. The combination of dilated convolution and channel attention block enhances feature representations. A new patch-wise regression loss (PRL) is proposed to improve the original pixel-wise loss for better convergence. Experimental results demonstrate the superiority of STDNet over state-of-the-art methods in video-based crowd counting.