【行人重识别最新进展】2017-2018年行人重识别算法精度统计

2017-2018年行人重识别算法在开源数据集上的表现

Market-1501

Method	Time	Single Query		Multi. Query
		rank-1	mAP	rank-1	mAP
Verif-Identif. + LSRO [1]	2017 ICCV	83.97	66.07	88.42	76.10
Basel. + LSRO [1]	2017 ICCV	78.06	56.23	85.12	68.52
SVDNet(C) [2]	2017 ICCV	80.5	55.9
SVDNet(R,1024-dim) [2]	2017 ICCV	82.3	62.1
M-net [3]	2017 ICCV	73.1
HP-net [3]	2017 ICCV	76.9
CADL [7]	2017 CVPR	73.84	47.11	80.85	55.58
Fusion [9]	2017 CVPR	80.31	57.53	86.79	66.70
SSM [10]	2017 CVPR	82.21	68.80	88.18	76.18
Spindle [12]	2017 CVPR	76.9
DeepAlign. [13]	2017 ICCV	81.0	63.4
Zhong et al. [14]	2017 CVPR	77.11	63.63
TriNet (Re- ranked) [15]	2017 ICCV	86.67	81.07	91.75	87.18
PDC [16]	2017 ICCV	84.14	63.41
DPFL [17]	2017 ICCV	88.6	72.6	92.2	80.4
DPFL(2+) [17]	2017 ICCV	88.9	73.1	92.3	80.7
PSE [18]	2018 CVPR	87.7	69.0
PSE+ ECN (rank-dist) [18]	2018 CVPR	90.3	84.0
IDE∗+Cam Style+RE [19]	2018 CVPR	89.49	71.55
MobileNet+DML [20]	2018 CVPR	87.73	68.83	91.66	77.14
Resnet50-res5c (multi-scale) 83.58 64.25+DSR (fine-tuning model) [21]	2018 CVPR	83.58	64.25
DuATM [23]	2018 CVPR	91.42	76.62
HA-CNN [25]	2018 CVPR	91.2	75.7	93.8	82.8
SPReIDcombined-ft*+re-ranking [26]	2018 CVPR	94.63	90.96
MLFN [27]	2018 CVPR	90.0	74.3	92.3	82.4
BraidNet-CS + SRL [29]	2018 CVPR	83.70	69.48
Pose-transfer (D, Tri) [30]	2018 CVPR	87.65	68.92
DaRe(De)+RE+RR [31]	2018 CVPR	90.9	86.7
TFusion-sup [32]	2018 CVPR	73.13
Chen et al. [33]	2018 ECCV	81.8	93.3	87.9	95.3
HAP2S_E [34]	2018 ECCV	84.20	69.76
HAP2S_P [34]	2018 ECCV	84.59	69.43
Mancs [35]	2018 ECCV	93.1	82.3	95.4	87.5
Suh et al. [37]	2018 ECCV	93.4	89.9	95.4	93.1
Base Model + SGGNN [38]	2018 ECCV	92.3	82.8

 

 

DukeMTMC-reID

Method	Time	rank-1	mAP
Basel. + LSRO [1]	2017 ICCV	67.68	47.13
SVDNet (C) [2]	2017 ICCV	67.6	45.8
SVDNet (R) [2]	2017 ICCV	76.7	56.8
DPFL [17]	2017 ICCV	79.2	60.6
PSE [18]	2018 CVPR	79.8	62.0
PSE+ ECN (rank-dist) [18]	2018 CVPR	85.2	79.8
IDE∗+Cam Style+RE [19]	2018 CVPR	78.32	57.61
DuATM [23]	2018 CVPR	81.82	64.58
HA-CNN [25]	2018 CVPR	80.5	63.8
Inception-V3ft*+re-ranking [26]	2018 CVPR	89.41	84.82
SPReIDcombined-ft*+re-ranking [26]	2018 CVPR	88.96	84.99
MLFN [27]	2018 CVPR	81.0	62.8
BraidNet-CS + SRL [29]	2018 CVPR	76.44	59.49
DaRe(De)+RE+RR [31]	2018 CVPR	84.4	80.0
HAP2S_E [34]	2018 ECCV	76.08	59.58
HAP2S_P [34]	2018 ECCV	75.94	60.64
Mancs [35]	2018 ECCV	84.9	71.8
Suh et al. [37]	2018 ECCV	88.3	83.9
Base Model + SGGNN [38]	2018 ECCV	81.1	68.2

 

 

CUHK03-labeled

Method	Time	rank-1	rank-5	rank-10	rank-20	mAP
Basel.+LSRO [1]	2017 ICCV	73.1	92.7	96.7		77.4
Verif-Identif. + LSRO [1]	2017 ICCV	84.6	97.6	98.9		87.4
SVDNet(C) [2]	2017 ICCV	68.5				73.3
SVDNet(R,1024-dim) [2]	2017 ICCV	81.8				84.8
M-net [3]	2017 ICCV	88.2	98.2	99.1	99.5
HP-net [3]	2017 ICCV	91.8	98.4	99.1	99.6
Quadruplet + MargOHNM [6]	2017 CVPR	75.53	95.15	99.16
Quadruplet [6]	2017 CVPR	74.47	96.92	98.95
Fusion [9]	2017 CVPR	74.21	94.33	97.54	99.25
SSM [10]	2017 CVPR	76.6		94.6	98.0
Spindle [12]	2017 CVPR	88.5	97.8	98.6	99.2
DeepAlign. [13]	2017 ICCV	85.4	97.6	99.4	99.9	90.9
PDC [16]	2017 ICCV	88.70	98.61	99.24	99.67
DPFL [17]	2017 ICCV	86.7	82.8
Guo et al. [24]	2018 CVPR	87.50	97.85	99.45
SPReIDcombined-ft*+re-ranking [26]	2018 CVPR	96.22	99.34	99.7
MLFN [27]	2018 CVPR	82.8
BraidNet-CS + SRL [29]	2018 CVPR	88.18		98.66	99.48
DaRe(De)+RE+RR [31]	2018 CVPR	73.8				74.7
Chen et al. [33]	2018 ECCV	92.5	98.8
HAP2S_P [34]	2018 ECCV	90.39	99.54	99.90
Mancs [35]	2018 ECCV	93.8	99.3	99.8
Suh et al. [37]	2018 ECCV	91.5	99.0	99.5	99.9
Base Model + SGGNN [38]	2018 ECCV	95.3	99.1	99.6		94.3
MC-PPMN (hnm) [39]	2018 AAAI	86.36	98.54	99.66

 

 

CUHK01(p=486)

Method	Time	rank-1	rank-5	rank-10	rank-20
Quadruplet + MargOHNM [6]	2017 CVPR	62.55	83.44	89.71
CSBT [8]	2017 CVPR	51.2	76.3		91.8
Spindle [12]	2017 CVPR	79.9	94.4	97.1	98.6
DeepAlign. [13]	2017 ICCV	75.0	93.5	95.7	97.7
Chen et al. [33]	2018 ECCV	84.8	95.1	98.4
Suh et al. [37]	2018 ECCV	80.7	94.4	97.3	98.6
MC-PPMN (hnm) [39]	2018 AAAI	78.95	94.67	97.64

 

 

CUHK01(p=100)

Method	Time	rank-1	rank-5	rank-10	rank-20
DeepAlign. [13]	2017 ICCV	88.5	98.4	99.6	99.9
Guo et al. [24]	2018 CVPR	88.20	98.20	99.35
BraidNet-CS + SRL [29]	2018 CVPR	93.04		99.97	100.00
Suh et al. [37]	2018 ECCV	90.4	97.1	98.1	98.9
MC-PPMN (hnm) [39]	2018 AAAI	93.45	99.62	99.98

 

 

Viper

Method	Time	rank-1	rank-5	rank-10	rank-20
M-net [3]	2017 ICCV	51.6	73.1	81.6	88.3
HP-net [3]	2017 ICCV	76.9	91.3	94.5	96.7
SHaPE [5]	2017 ICCV	34.26	57.34	67.86	80.78
Quadruplet + MargOHNM [6]	2017 CVPR	49.05	73.10	81.96
CSBT [8]	2017 CVPR	36.6	66.2		88.3
Fusion [9]	2017 CVPR	38.08	64.14	73.52	82.91
SSM [10]	2017 CVPR	53.73		91.49	96.08
Spindle [12]	2017 CVPR	53.8	74.1	83.2	92.1
DeepAlign. [13]	2017 ICCV	48.7	74.7	85.1	93.0
PDC [16]	2017 ICCV	51.27	74.05	84.18	91.46
Guo et al. [24]	2018 CVPR	50.10	73.10	84.35
MC-PPMN [39]	2018 AAAI	50.13	81.17	91.46

 

 

PRW

Method	Time	rank-1	mAP
NPSM [4]	2017 ICCV	53.1	24.2
Zhong et al. [14]	2017 CVPR	52.54	31.51

 

 

MARS

Method	time	Single Query
		rank-1	rank-5	rank-20	mAP
Fusion+XQDA [9]	2017 CVPR	71.77	86.57	93.08	56.05
STRN [11]	2017 CVPR	70.6	90.0	97.6	50.7
Zhong et al. [14]	2017 CVPR	73.94			68.45
TriNet (Re- ranked) [15]	2017 ICCV	81.21	90.76		77.43
PSE [18]	2018 CVPR	72.1			56.9
PSE+ ECN (rank-dist) [18]	2018 CVPR	76.7			71.8
SpaAtn+Q+TemAtn+Ind [22]	2018 CVPR	82.3			65.8
DuATM [23]	2018 CVPR	78.74	90.86	95.76	62.26
Zhang et al. [28]	2018 CVPR	71.2	85.7	91.8	94.3
DaRe(De)+RE+RR [31]	2018 CVPR	85.1			81.9
Suh et al. [37]	2018 ECCV	85.1	94.2	97.4	83.9

 

 

QMUL GRID

Method	Time	rank-1	rank-5	rank-10	rank-20
SSM[ [10]	2017 CVPR	27.20		61.12	70.56
TFusion-sup [32]	2018 CVPR	64.10	91.90	96.50
NK3ML [36]	2018 ECCV	27.20		60.96	71.04

 

 

iLIDS-VID

Method	Time	rank-1	rank-5	rank-10	rank-20
STRN [11]	2017 CVPR	55.2		86.5	97.0
Spindle [12]	2017 CVPR	66.3	86.6	91.8	95.3
SpaAtn+Q+TemAtn+Ind [22]	2018 CVPR	80.2
Zhang et al. [28]	2018 CVPR	60.2	84.7	91.7	95.2
MC-PPMN [39]	2018 AAAI	62.69	84.80	93.31

 

 

PRID2011

Method	Time	rank-1	rank-5	rank-10	rank-20
SSM [10]	2017 CVPR	72.98		96.76	99.11
STRN [11]	2017 CVPR	79.4		94.4	99.3
Spindle [12]	2017 CVPR	67.0	89.0	89.0	92.0
SpaAtn+Q+TemAtn+Ind [22]	2018 CVPR	93.2
Zhang et al. [28]	2018 CVPR	85.2	97.1	98.9	99.6
MC-PPMN [39]	2018 AAAI	34.00	60.00	69.00

 

 

3DPeS

Method	Time	rank-1	rank-5	rank-10	rank-20
Spindle [12]	2017 CVPR	62.1	83.4	90.5	95.7

 

参考文献：

[1] Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in vitro

[2] SVDNet for Pedestrian Retrieval

[3] HydraPlus-Net: Attentive Deep Features for Pedestrian Analysis

[4] Neural Person Search Machines

[5] SHaPE: A Novel Graph Theoretic Algorithm for Making Consensus-based Decisions in Person Re-identification Systems

[6] Beyond triplet loss: a deep quadruplet network for person re-identification

[7] Consistent-Aware Deep Learning for Person Re-identification in a Camera Network

[8] Fast Person Re-identification via Cross-camera Semantic Binary Transformation

[9] Learning Deep Context-aware Features over Body and Latent Parts for Person Reidentification

[10] Scalable Person Re-identification on Supervised Smoothed Manifold

[11] see the forest for the trees：spitial and temporal recurrent neural networks for video-based re-id

[12] Spindle Net: Person Re-identification with Human Body Region Guided Feature Decomposition and Fusion

[13] Deeply-Learned Part-Aligned Representations for Person Re-Identification

[14] Re-ranking Person Re-identification with k-reciprocal Encoding

[15] In Defense of the Triplet Loss for Person Re-Identification

[16] Pose-driven Deep Convolutional Model for Person Re-identification

[17] Person Re-Identification by Deep Learning Multi-Scale Representations

[18] A Pose-Sensitive Embedding for Person Re-Identification with Expanded Cross Neighborhood Re-Ranking

[19] Camera Style Adaptation for Person Re-identification

[20] Deep Mutual Learning

[21] Deep Spatial Feature Reconstruction for Partial Person Re-identification: Alignment-Free Approach

[22] Diversity Regularized Spatiotemporal Attention for Video-based Person Re-identification

[23] Dual Attention Matching Network for Context-Aware Feature Sequence based Person Re-Identification

[24] Efficient and Deep Person Re-Identification using Multi-Level Similarity

[25] Harmonious Attention Network for Person Re-Identification

[26] Human Semantic Parsing for Person Re-identification

[27] Multi-Level Factorisation Net for Person Re-Identification

[28] Multi-shot Pedestrian Re-identification via Sequential Decision Making

[29] Person Re-identification with Cascaded Pairwise Convolutions

[30] Pose Transferrable Person Re-Identification

[31] Resource Aware Person Re-identification across Multiple Resolutions

[32] Unsupervised Cross-dataset Person Re-identification by Transfer Learning of Spatial-Temporal Patterns

[33] Improving Deep Visual Representation for Person Re-identification by Global and Local Image-language Association

[34] Hard-Aware Point-to-Set Deep Metric for Person Re-identification

[35] Mancs: A Multi-task Attentional Network with Curriculum Sampling for Person Re-identification

[36] Maximum Margin Metric Learning Over Discriminative Nullspace for Person Re-identification

[37] Part-Aligned Bilinear Representations for Person Re-identification

[38] Person Re-identification with Deep Similarity-Guided Graph Neural Network

[39] Multi-Channel Pyramid Person Matching Network for Person Re-Identification