Posture-Guided and Multi-Granularity Feature Fusion for Person Reidentification

Method	Rank-1 /%	mAP /%
BoW+Kissme^[17]	25.13	12.17
LOMO+XQDA^[1]	30.75	17.04
PAN^[19]	71.59	51.51
SVDNet^[18]	76.7	56.8
AACN^[20]	76.84	59.25
PSE^[22]	79.8	62.0
HA-CNN^[24]	80.5	63.8
MLFN^[26]	81.2	62.8
PCB^[8]	83.3	69.2
Part-aligned^[23]	84.4	69.3
Ours(G)	83.7	70.1

Table 1. Details of the datasets
View table
Table 1. Details of the datasets
Dataset ID Train ID Test ID Bbox Bbox /ID Cam
Market1501 1501 751 750 32668 21.8 6
CUHK03-NP 1467 767 700 14097 9.6 2
DukeMTMC-reID 1404 702 702 36411 25.9 8

Table 2. Experimental data of M=4 and descriptor H in Market1501
View table
Table 2. Experimental data of M=4 and descriptor H in Market1501
w×h Rank-1 /% Rank-5 /% Rank-10 /% mAP /%
64×192 84.4 93.4 95.5 66.3
96×288 87.4 95.1 96.9 71.3
128×384 89.0 95.8 97.6 73.6
160×480 90.3 96.5 97.6 74.6
192×576 90.2 96.6 97.8 74.3

Table 3. Experimental data of M=6 and descriptor H in Market1501
View table
Table 3. Experimental data of M=6 and descriptor H in Market1501
w×h Rank-1 /% Rank-5 /% Rank-10 /% mAP /%
64×192 85.3 93.9 96.3 68.8
96×288 88.4 95.2 97.0 72.9
128×384 89.7 95.5 96.9 73.8
160×480 91.0 96.1 97.7 74.8
192×576 90.3 96.4 97.6 74.1

Table 4. Experimental data of M=6 and descriptor G in Market1501
View table
Table 4. Experimental data of M=6 and descriptor G in Market1501
w×h Rank-1 /% Rank-5 /% Rank-10 /% mAP /%
64×192 87.7 94.9 96.4 69.7
96×288 91.5 96.6 97.6 75.8
128×384 93.2 97.1 98.1 78.3
160×480 92.4 97.0 97.8 77.7
192×576 92.5 97.3 98.4 77.2

Table 5. Experimental data of M=4 and descriptor G in CUHK03-NP
View table
Table 5. Experimental data of M=4 and descriptor G in CUHK03-NP
w×h Rank-1 /% Rank-5 /% Rank-10 /% mAP /%
64×192 47.9 68.4 76.6 47.5
96×288 52.8 73.1 81.6 52.1
128×384 59.3 78.4 83.9 57.2
160×480 60.4 77.7 84.6 58.1
192×576 51.3 72.2 79.6 50.5

Table 6. Experimental data of M=6 and descriptor G in CUHK03-NP
View table
Table 6. Experimental data of M=6 and descriptor G in CUHK03-NP
w×h Rank-1 /% Rank-5 /% Rank-10 /% mAP /%
64×192 47.8 68.8 77.5 47.1
96×288 55.3 74.5 82.3 54.5
128×384 59.7 77.1 84.3 57.7
160×480 56.7 75.1 82.4 54.3
192×576 56.4 73.9 82.5 54.1

Table 7. Experimental data of M=6 and descriptor G in DukeMTMC-reID
View table
Table 7. Experimental data of M=6 and descriptor G in DukeMTMC-reID
w×h Rank-1/% Rank-5/% Rank-10/% mAP/%
64×192 79.7 88.5 91.1 61.6
96×288 83.6 92.0 93.6 68.5
128×384 83.7 92.1 94.1 70.1
160×480 83.6 91.7 93.7 70.0
192×576 81.2 90.6 92.7 68.5

Table 8. Comparison ofmethods in Market1501 dataset

View table

Table 8. Comparison ofmethods in Market1501 dataset

Method	Rank-1 /%	mAP /%
BoW+Kissme^[17]	20.76	44.42
LOMO+XQDA^[1]	72.58	51.96
SVDNet^[18]	82.3	62.1
PAN ^[19]	82.8	63.4
AACN^[20]	85.90	66.87
AOS^[21]	86.5	70.4
PSE^[22]	87.7	69.0
Part-aligned^[23]	88.8	74.5
HA-CNN^[24]	91.2	75.7
PCB^[8]	92.4	77.3
Ours(H)	91.0	74.8
Ours(G)	93.2	78.3

Table 9. Comparison ofmethods in CUHK03-NP dataset

View table

Table 9. Comparison ofmethods in CUHK03-NP dataset

Method	Rank-1 /%	mAP /%
BoW+Kissme^[17]	6.4	6.4
LOMO+XQDA^[1]	12.8	11.5
PAN^[19]	36.3	34.0
MutiScale^[25]	40.7	37.0
SVDNet^[18]	41.5	37.3
HA-CNN^[24]	41.7	38.6
AOS^[21]	47.1	43.3
MLFN^[26]	52.8	47.8
DaRe^[27]	55.1	51.3
PCB^[8]	61.3	54.2
Ours(G)	60.4	58.1

Tools

Get Citation

Copy Citation Text

Liang Zhang, Jin Che. Posture-Guided and Multi-Granularity Feature Fusion for Person Reidentification[J]. Laser & Optoelectronics Progress, 2019, 56(20): 201501

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites