1.Person Re-identification by Video Ranking --ECCV2014 [paper ]
特征(feature): HOG3D
本文的主要贡献&创新点:开源了新的视频序列数据集iLIDS-VID;能从有噪声的帧序列中选出关键帧;学习一个视频排序函数.
实验结果(results):
iLIDS-VID PRID
Rank1 28.9 23.3
1.Sparse Re-ID: Block Sparsity for Person Re-identification --CVPR2015 [paper ]
特征(feature): Color Histograms Schmid & Gabor Filters
实验结果(results):
iLIDS-VID PRID
Rank1 24.9 35.1
2.A Spatio-temporal Appearance Representation for Video-based Pedestrian Re-identification --ICCV2015 [paper ]
特征(feature): Fiser vector
实验结果(results):
iLIDS-VID PRID
Rank1 44.3 64.1
1.Deep Recurrent Convolutional Networks for Video-based Person Re-identification: An End-to-End Approach --arxiv2016 [paper ]
特征(feature): four CNN(四层卷积神经网络)
实验结果(results):
iLIDS-VID PRID
Rank1 42.6 49.8
2.Top-push Video-based Person Re-identification --CVPR 2016 [paper ]
特征(feature): HOG3D+color histograms+LBP
实验结果(results):
iLIDS-VID PRID
Rank1 56.33 56.74
3.Person Re-identification by Exploiting Spatio-temporal Cues and Multi-view Metric Learning --IEEE SRL 2016 [paper ]
特征(feature): LBP
实验结果(results):
iLIDS-VID PRID
Rank1 69.13 66.78
4.Person Re-identification via Recurrent Feature Aggregation --ECCV2016 [paper ][code ]
特征(feature): LBP+HSV+lab color channels
实验结果(results):
iLIDS-VID PRID
Rank1 49.3 64.1
5.Recurrent Convolutional Network for Video-based Person Re-identification --CVPR2016 [paper ] [code ]博客解读:[CSDN ]
特征(feature): CNN + RNN (非常经典的结构)
实验结果(results):
iLIDS-VID PRID
Rank1 58 70
6.Temporally aligned pooling representation for video-based person re-identification --ICIP2016 [paper ]
特征(feature): Local maximal occurrence representation (LOMO) 时间对齐池化表征(与众不同,关注步态周期循环)
实验结果(results):
iLIDS-VID PRID
Rank1 55 68.6
1.Video-based Person Re-identification with Accumulative Motion Contex --TCSVT2017 [paper ]
特征(feature): CNN + RNN + opticflow
实验结果(results):
iLIDS-VID PRID
Rank1 65.3 78
2.Learning Compact Appearance Representation for Video-based Person Re-identification --arxiv2017 [paper ]
特征(feature): Five CNN (5层卷积神经网络)
实验结果(results):
iLIDS-VID PRID
Rank1 60.4 83.3
3.See the Forest for the Trees: Joint Spatial and Temporal Recurrent Neural Networks for Video-based Person Re-identification --CVPR2017 [paper ]
特征(feature): Temporal Attention Model (TAM) + Spatial Recurrent Model (SRM)
实验结果(results):
iLIDS-VID PRID Mars
Rank1 55.2 79.4 70.6
mAP -- -- 50.7
4.Quality Aware Network for Set to Set Recognition --CVPR2017 [paper ] [code ]
特征(feature): CNN(GoogleNet)
实验结果(results):
iLIDS-VID PRID
Rank1 68.0 90.3
5.Jointly Attentive Spatial-Temporal Pooling Networks for Video-based Person Re-identification --ICCV2017 [paper ] [code ]博客解读:[CSDN ]
特征(feature): (3层)CNN + RNN + attention
实验结果(results):
iLIDS-VID PRID Mars
Rank1 62 77 44
6.A Two Stream Siamese Convolutional Neural Network For Person Re-identification --ICCV2017 [paper ]
特征(feature): CNN + RNN + attention => siamese network
实验结果(results):
iLIDS-VID PRID
Rank1 60 78
7.Deep Cross-Modality Alignmeant for Multi-Shot Person Re-Identification --MM2017 [paper ]
特征(feature): (3层)CNN + RNN + avgpooling
实验结果(results):
iLIDS-VID PRID Mars
Rank1 60 80 63
8.Data Generation for Improving Person Re-identification --MM2017 [paper ]
特征(feature): (3层)CNN + RNN + avgpooling
实验结果(results):
iLIDS-VID PRID
Rank1 66 79
9.Three-Stream Convolutional Networks for Video-based Person Re-Identification --arxiv2017 [paper ]
特征(feature): (4层)CNN + RNN + avgpooling
实验结果(results):
iLIDS-VID PRID Mars
Rank1 67.5 79.7 45.6
10.Phasic Maximal and Local Maximal Occurrence Representation for Video-Based Person Re-identification --ICCSN2017 [paper ]
特征(feature): Phasic Maximal and Local Maximal Occurrence (PM-LOMO)
实验结果(results):
iLIDS-VID PRID
Rank1 57 82.58
1.TRegion-based Quality Estimation Network for Large-scale Person Re-identification --AAAI2018 [paper ]
特征(feature): (4层)CNN + RNN + avgpooling
实验结果(results):
iLIDS-VID PRID Mars
Rank1 76.1 92.4 77.83
mAP -- -- 71.14
2.Video Person Re-identification by Temporal Residual Learning --arxiv2018 [paper ]
特征(feature): GoogleNet + BiLSTM
实验结果(results):
iLIDS-VID PRID Mars
Rank1 57.7 87.8 79.3
3.Diversity Regularized Spatiotemporal Attention for Video-based Person Re-identification --CVPR2018 [paper ] [code ]
特征(feature): Resnet50 + attention
实验结果(results):
iLIDS-VID PRID Mars
Rank1 80.2 93.2 82.3
mAP -- -- 65.8
4.Multi-shot Pedestrian Re-identification via Sequential Decision Making --CVPR2018 [paper ] [code ]
特征(feature): Inception-BN or AlexNet
实验结果(results):
iLIDS-VID PRID Mars
Rank1 60.2 85.2 71.2
5.Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning --CVPR2018 [paper ] [code ]
特征(feature): ResNet50 + avgpooling
实验结果(results):
iLIDS-VID PRID Mars DukeMTMC-VideoReID
Rank1 -- -- 62.67 72.79
mAP -- -- 42.45 63.23
6.Video Person Re-identification with Competitive Snippet-similarity Aggregation and Co-attentive Snippet Embedding --CVPR2018 [paper ] [code ]
特征(feature): ResNet50 + attention
实验结果(results):
iLIDS-VID PRID Mars
Rank1 85.4 93.0 86.3
mAP 87.8 94.5 76.1
7.SCAN: Self-and-Collaborative Attention Network for Video Person Re-identification --CVPR2018 [paper ]
特征(feature): ResNet50 + attention
实验结果(results):
iLIDS-VID PRID Mars
Rank1 88.0 95.3 87.2
mAP 89.9 95.8 77.2
8.Video-based Person Re-identification via 3D Convolutional Networks and Non-local Attention --arxiv2018 [paper ]
特征(feature): ResNet50-3D + avgpooling
实验结果(results):
iLIDS-VID PRID Mars
Rank1 81.3 91.2 84.3
mAP -- -- 77
9.Spatial-Temporal Synergic Residual Learning for Video Person Re-Identification --arxiv2018 [paper ]
特征(feature): CNN + RNN + avgpooling
实验结果(results):
iLIDS-VID PRID Mars
Rank1 70 88 76.7