Convolutional Neural Network Method for Crowd Counting Improved using Involution Operator

Table 1. Value of coefficient α in different datasets
View table
Table 1. Value of coefficient α in different datasets
Dataset $α$
SHHA 1000
SHHB 100
UCF-QNRF 1000
UCF_CC_50 100

Table 2. Comparison results of the different methods

View table

Table 2. Comparison results of the different methods

Method	SHHA		SHHB		UCF-QNRF		UCF_CC_50
Method	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE
MCNN^［24］	110.2	173.2	26.4	41.3	243.5	364.7	467.0	498.5
Switching CNN^［26］	90.4	135.0	21.6	33.4	228.0	445.0	318.1	439.2
CMTL^［31］	101.3	152.4	20.0	31.1	252	514	322.8	397.9
CSRNet^［16］	68.2	115.0	10.6	16.0	120.3	208.5	266.1	397.5
SANet^［32］	67.0	104.5	8.4	13.6	-	-	-	-
ACSPNet^［18］	85.2	137.1	15.4	23.1	-	-	-	-
PCCNet^［10］	73.5	124.0	11.0	19.0	149.0	247.0	240.0	315.5
AMCNN^［2］	76.1	110.7	15.3	27.4	-	-	-	-
LSC-CNN^［33］	66.4	117.0	8.1	12.7	120.5	218.2	255.6	302.7
TEDNet^［30］	64.2	109.1	8.2	12.8	113.0	188.0	249.4	354.5
SCLNet^［20］	67.9	102.9	9.1	14.1	109.6	182.5	258.9	326.2
DSPNet^［13］	68.2	107.8	8.9	14.0	107.5	182.7	243.3	307.6
MSCANet^［34］	66.5	109.4	-	-	104.1	183.8	242.8	329.8
Method in Ref.［15］	61.9	100.5	7.4	11.7	104.8	182.3	212.3	289.6
AMS-Net^［11］	63.8	108.5	7.3	11.8	86.5	167.2	236.5	319.2
Proposed method	61.1	101.3	7.0	11.3	102.5	181.7	202.0	288.7

Table 3. Ablation study on the SHHA dataset

View table

Table 3. Ablation study on the SHHA dataset

Component	MAE	MSE	Time /ms	Parameters /10⁶
VGG（dilation rate is 2）（baseline）	69.1	106.2	166.54	11.54
VGG+1INV（dilation rate is 2）	66.9	105.2	145.38	11.81
VGG+2INV（dilation rate is 2）	64.9	101.4	133.85	12.08
VGG+3INV（dilation rate is 1）	67.0	108.5	132.20	12.35
VGG+3INV（dilation rate is 2）	63.6	103.6	127.53	12.35
VGG+3INV（dilation rate is 2）+residual connection	62.9	106.5	133.63	12.35
VGG+3INV（dilation rate is 2）+residual connection+Loss	61.1	101.3	136.43	12.35

Tools

Get Citation

Copy Citation Text

Zhaoxin Li, Shuhua Lu, Lingqiang Lan, Qiyuan Liu. Convolutional Neural Network Method for Crowd Counting Improved using Involution Operator[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1815004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites