MSIANet：多尺度交互注意力人群计数网络

张世辉; 赵维勃; 王磊; 王威; 李群鹏

doi:10.11999/JEIT220644

MSIANet：多尺度交互注意力人群计数网络

doi: 10.11999/JEIT220644

张世辉^{1, 2},
赵维勃^1, ,,
王磊¹,
王威¹,
李群鹏¹

1.
燕山大学信息科学与工程学院秦皇岛 066004
2.
河北省计算机虚拟技术与系统集成重点实验室秦皇岛 066004

基金项目: 中央引导地方科技发展资金项目(216Z0301G)，河北省自然科学基金(F2019203285)，河北省创新能力提升计划项目(22567626H)

详细信息

作者简介:
张世辉：男，教授，博士生导师，研究方向为视觉信息处理、模式识别

赵维勃：男，硕士生，研究方向为人群计数、计算机视觉

王磊：男，博士生，研究方向为草图识别、计算机视觉

王威：男，硕士生，研究方向为人群计数、计算机视觉

李群鹏：男，硕士生，研究方向为人群计数、计算机视觉

通讯作者:
赵维勃　zhaowb@stumail.ysu.edu.cn

中图分类号: TN911.73; TP391.41
计量
- 文章访问数: 457
- HTML全文浏览量: 549
- PDF下载量: 110
- 被引次数: 0
出版历程
- 收稿日期: 2022-05-19
- 修回日期: 2022-07-29
- 网络出版日期: 2022-08-22
- 刊出日期: 2023-06-10

MSIANet: Multi-scale Interactive Attention Crowd Counting Network

ZHANG Shihui^{1, 2},
ZHAO Weibo^{1
, ,},
WANG Lei¹,
WANG Wei¹,
LI Qunpeng¹

1.
School of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, China
2.
The Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province, Qinhuangdao 066004, China

Funds: The Central Government Guided Local Funds for Science and Technology Development (216Z0301G), The Natural Science Foundation of Hebei Province in China (F2019203285), The Innovation Capability Improvement Plan Project of Hebei Province (22567626H)

摘要

摘要: 尺度变化、遮挡和复杂背景等因素使得拥挤场景下的人群数量估计成为一项具有挑战性的任务。为了应对人群图像中的尺度变化和现有多列网络中规模限制及特征相似性问题，该文提出一种多尺度交互注意力人群计数网络(Multi-Scale Interactive Attention crowd counting Network, MSIANet)。首先，设计了一个多尺度注意力模块，该模块使用4个具有不同感受野的分支提取不同尺度的特征，并将各分支提取的尺度特征进行交互，同时，使用注意力机制来限制多列网络的特征相似性问题。其次，在多尺度注意力模块的基础上设计了一个语义信息融合模块，该模块将主干网络的不同层次的语义信息进行交互，并将多尺度注意力模块分层堆叠，以充分利用多层语义信息。最后，基于多尺度注意力模块和语义信息融合模块构建了多尺度交互注意力人群计数网络，该网络充分利用多层次语义信息和多尺度信息生成高质量人群密度图。实验结果表明，与现有代表性的人群计数方法相比，该文提出的MSIANet可有效提升人群计数任务的准确性和鲁棒性。
- 人群计数 /
- 估计密度图 /
- 注意力机制 /
- 多尺度特征
Abstract: Factors such as scale variation, occlusion and complex backgrounds make crowd number estimation in crowded scenes a challenging task. To cope with the scale variation in crowd images and the scope limitation and the feature similarity problem in existing multi-column networks, a Multi-Scale Interactive Attention crowd counting Network (MSIANet) is proposed in this paper. Firstly, a multi-scale attention module is designed, which uses four branches with different perceptual fields to extract features at different scales and interacts the scale features extracted from each branch. At the same time, an attention mechanism is used to limit the feature similarity problem of the multi-column network. Secondly, a semantic information fusion module is designed based on the multi-scale attention module, which interacts different levels of semantic information of the backbone network and stacks the multi-scale attention module in layers to make full use of the multi-layer semantic information. Finally, a multi-scale interactive attention crowd counting network is constructed based on the multi-scale attention module and the semantic information fusion module, which makes full use of multi-level semantic information and multi-scale information to generate high-quality crowd density maps. The experimental results show that compared with the existing representative crowd counting methods, the proposed MSIANet can effectively improve the accuracy and robustness of the crowd counting task.
- Crowd counting /
- Estimated density map /
- Attention mechanism /
- Multi-scale features

HTML全文

图 1 多尺度交互注意力人群计数网络

下载: 全尺寸图片幻灯片

图 2 多尺度注意力模块

下载: 全尺寸图片幻灯片

图 3 多尺度交互结构

下载: 全尺寸图片幻灯片

图 4 全局空间注意力机制

下载: 全尺寸图片幻灯片

图 5 全局通道注意力机制

下载: 全尺寸图片幻灯片

图 6 ShanghaiTech Part A数据集密度图可视化结果

下载: 全尺寸图片幻灯片

图 7 ShanghaiTech Part B数据集密度图可视化结果

下载: 全尺寸图片幻灯片

图 8 UCF_QNRF数据集密度图可视化结果

下载: 全尺寸图片幻灯片

图 9 UCF_CC_50数据集密度图可视化结果

下载: 全尺寸图片幻灯片

图 10 ShanghaiTech Part A上估计人数与真实人数比较

下载: 全尺寸图片幻灯片

图 11 ShanghaiTech Part B上估计人数与真实人数比较

下载: 全尺寸图片幻灯片

图 12 UCF_QNRF上估计人数与真实人数比较

下载: 全尺寸图片幻灯片

图 13 UCF_CC_50上估计人数与真实人数比较

下载: 全尺寸图片幻灯片

表 1 在3个人群计数基准数据集上使用MAE和RMSE指标进行评估(加粗表示最好结果)

方法	ShanghaiTech A		ShanghaiTech B		UCF_QNRF		UCF_CC_50
方法	MAE	RMSE	MAE	RMSE	MAE	RMSE	MAE	RMSE
MCNN^[12] (2016)	110.2	173.2	26.4	41.3	277.0	426.0	377.6	509.1
SANet ^[13] (2018)	67.0	104.5	8.4	13.6	–	–	258.4	334.9
CSRNet ^[7] (2018)	68.2	115.0	10.6	16.0	–	–	266.1	397.5
Switch-CNN^[14] (2017)	90.4	135.0	21.6	33.4	228.0	445.0	318.1	439.2
ADCrowdNet^[19] (2019)	63.2	98.9	8.2	15.7	–	–	266.4	358.0
TEDNet^[15] (2019)	64.2	109.1	8.2	12.8	113.0	188.0	249.4	354.5
EPA^[16] (2020)	60.9	91.6	7.9	11.6	–	–	205.1	342.1
DUBNet^[8] (2020)	64.6	106.8	7.7	12.5	105.6	180.5	243.8	329.3
DPDNet^[17] (2021)	66.6	120.3	7.9	12.4	126.8	208.6	–	–
MLAttnCNN^[20] (2021)	–	–	7.5	11.6	101.0	175.0	200.8	273.8
URC^[9] (2021)	72.8	111.6	12.0	18.7	128.1	218.0	293.9	443.0
MPS^[18] (2022)	71.1	110.7	9.6	15.0	–	–	–	–
AutoScale^[10] (2022)	65.8	112.1	8.6	13.9	104.4	174.2	–	–
FusionCount^[11] (2022)	62.2	101.2	6.9	11.8	–	-	–	–
MSIANet(本文)	55.6	99.2	6.6	11.0	94.8	184.6	194.5	273.3

下载: 导出CSV

表 2 消融实验结果

变体模型	MAE	RMSE
MSIANet的前端网络+后端网络	63.4	105.2
MSIANet的前端网络+MASM+后端网络	58.5	101.8
MSIANet w/o MSAM	60.7	101.0
MSIANet w/o GCAM	57.3	100.6
MSIANet w/o GSAM	57.1	99.5
MSIANet	55.6	99.2

下载: 导出CSV

参考文献(23)

[1]	徐涛, 段仪浓, 杜佳浩, 等. 基于多尺度增强网络的人群计数方法[J]. 电子与信息学报, 2021, 43(6): 1764–1771. doi: 10.11999/JEIT200331 XU Tao, DUAN Yinong, DU Jiahao, et al. Crowd counting method based on multi-scale enhanced network[J]. Journal of Electronics &Information Technology, 2021, 43(6): 1764–1771. doi: 10.11999/JEIT200331
[2]	万洪林, 王晓敏, 彭振伟, 等. 基于新型多尺度注意力机制的密集人群计数算法[J]. 电子与信息学报, 2022, 44(3): 1129–1136. doi: 10.11999/JEIT210163 WAN Honglin, WANG Xiaomin, PENG Zhenwei, et al. Dense crowd counting algorithm based on new multi-scale attention mechanism[J]. Journal of Electronics &Information Technology, 2022, 44(3): 1129–1136. doi: 10.11999/JEIT210163
[3]	TOPKAYA I S, ERDOGAN H, and PORIKLI F. Counting people by clustering person detector outputs[C]. Proceedings of the 11th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS), Seoul, Korea (South), 2014: 313–318.
[4]	LI Min, ZHANG Zhaoxiang, HUANG Kaiqi, et al. Estimating the number of people in crowded scenes by MID based foreground segmentation and head-shoulder detection[C]. Proceedings of the 19th International Conference on Pattern Recognition, Tampa, USA, 2008: 1–4.
[5]	CHAN A B, LIANG Z S J, and VASCONCELOS N. Privacy preserving crowd monitoring: Counting people without people models or tracking[C]. Proceedings of 2008 IEEE Conference on Computer Vision and Pattern Recognition, Anchorage, USA, 2008: 1–7.
[6]	CHEN Ke, LOY C C, GONG Shaogang, et al. Feature mining for localised crowd counting[C]. Proceedings of the British Machine Vision Conference, Surrey, UK, 2012: 21.1–21.11.
[7]	LI Yuhong, ZHANG Xiaofan, and CHEN Deming. CSRNet: Dilated convolutional neural networks for understanding the highly congested scenes[C]. Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 1091–1100.
[8]	OH M H, OLSEN P, and RAMAMURTHY K N. Crowd counting with decomposed uncertainty[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11799–11806. doi: 10.1609/aaai.v34i07.6852
[9]	XU Yanyu, ZHONG Ziming, LIAN Dongze, et al. Crowd counting with partial annotations in an image[C]. Proceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, Canada, 2021: 15550–15559.
[10]	XU Chenfeng, LIANG Dingkang, XU Yongchao, et al. AutoScale: Learning to scale for crowd counting[J]. International Journal of Computer Vision, 2022, 130(2): 405–434. doi: 10.1007/s11263-021-01542-z
[11]	MA Yiming, SANCHEZ V, and GUHA T. FusionCount: Efficient crowd counting via multiscale feature fusion[C]. Proceedings of the IEEE International Conference on Image Processing, Bordeaux, France, 2022.
[12]	ZHANG Yingying, ZHOU Desen, CHEN Siqin, et al. Single-image crowd counting via multi-column convolutional neural network[C]. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 589–597.
[13]	CAO Xinkun, WANG Zhipeng, ZHAO Yanyun, et al. Scale aggregation network for accurate and efficient crowd counting[C]. Proceedings of the 15th European Conference on Computer Vision, Munich, Germany, 2018: 757–773.
[14]	SAM D B, SURYA S, and BABU R V. Switching convolutional neural network for crowd counting[C]. Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, USA, 2017: 4031–4039.
[15]	JIANG Xiaolong, XIAO Zehao, ZHANG Baochang, et al. Crowd counting and density estimation by trellis encoder-decoder networks[C]. Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, USA, 2019: 6126–6135.
[16]	YANG Yifan, LI Guorong, DU Dawei, et al. Embedding perspective analysis into multi-column convolutional neural network for crowd counting[J]. IEEE Transactions on Image Processing, 2020, 30: 1395–1407. doi: 10.1109/TIP.2020.3043122
[17]	LIAN Dongze, CHEN Xianing, LI Jing, et al. Locating and counting heads in crowds with a depth prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, To be Published.
[18]	ZAND M, DAMIRCHI H, FARLEY A, et al. Multiscale crowd counting and localization by multitask point supervision[C]. Proceedings of 2022 IEEE International Conference on Acoustics, Speech and Signal Processing, Singapore, Singapore, 2022.
[19]	LIU Ning, LONG Yongchao, ZOU Changqing, et al. ADCrowdNet: An attention-injective deformable convolutional network for crowd understanding[C]. Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 3220–3229.
[20]	TIAN Mengxiao, GUO Hao, and LONG Chengjiang. Multi-level attentive convoluntional neural network for crowd counting[J]. arXiv: 2105.11422, 2021.
[21]	LIU Yichao, SHAO Zongru, and HOFFMANN N. Global attention mechanism: Retain information to enhance channel-spatial interactions[J]. arXiv: 2112.05561, 2021.
[22]	IDREES H, TAYYAB M, ATHREY K, et al. Composition loss for counting, density map estimation and localization in dense crowds[C]. Proceedings of the 15th European Conference on Computer Vision, Munich, Germany, 2018: 544–559.
[23]	IDREES H, SALEEMI I, SEIBERT C, et al. Multi-source multi-scale counting in extremely dense crowd images[C]. Proceedings of 2013 IEEE Conference on Computer Vision and Pattern Recognition, Portland, USA, 2013: 2547–2554.