This repository contains a curated list of person re-identification papers and BibTeX entries (mostly until 2021) orgnized by topics.
- (Book 14) Person Re-Identification
- (arXiv 16) Person Re-Identification: Past, Present and Future
- (TPAMI 18) A Systematic Evaluation and Benchmark for Person Re-Identification: Features, Metrics, and Datasets
- (TCSVT 19) A Survey of Open-World Person Re-Identification
- (IJCAI 20) Beyond Intra-Modality: A Survey of Heterogeneous Person Re-Identification
- (TPAMI 21) Deep Learning for Person Re-Identification: A Survey and Outlook
- (arXiv 22) Person Re-Identification: A Retrospective on Domain Specific Open Challenges and Future Trends
Verification
- (ICPR 14) Deep Metric Learning for Person Re-Identification
- (CVPR 15) An Improved Deep Learning Architecture for Person Re-Identification
- (ECCV 16) Gated Siamese Convolutional Neural Network Architecture for Human Re-Identification
- (ICCV 17) A Two Stream Siamese Convolutional Neural Network for Person Re-Identification
- (CVPR 22) Dual Cross-Attention Learning for Fine-Grained Visual Categorization and Object Re-Identification
Identification
- (CVPR 16) Learning a Discriminative Null Space for Person Re-Identification
- (CVPR 17) Re-ranking Person Re-Identification with
$k$ -reciprocal Encoding - (ICCV 17) SVDNet for Pedestrian Retrieval
- (arXiv 17) In Defense of the Triplet Loss for Person Re-Identification
- (TOMCCAP 17) A Discriminatively Learned CNN Embedding for Person Reidentification
- (CVPRW 19) Bag of Tricks and a Strong Baseline for Deep Person Re-Identification
- (ICCV 19) Omni-Scale Feature Learning for Person Re-Identification
- (ICCV 21) TransReID: Transformer-based Object Re-Identification
- (ACM MM 21) HAT: Hierarchical Aggregation Transformers for Person Re-identification
- (CVPR 22) NFormer: Robust Person Re-identification with Neighbor Transformer
Attention
- (CVPR 18) Harmonious Attention Network for Person Re-Identification
- (CVPR 19) Towards Rich Feature Discovery with Class Activation Maps Augmentation for Person Re-Identification
- (CVPR 19) Interaction-and-Aggregation Network for Person Re-Identification
- (ICCV 19) Mixed High-Order Attention Network for Person Re-Identification
- (ICCV 19) Second-Order Non-local Attention Networks for Person Re-Identification
- (ICCV 19) ABD-Net: Attentive but Diverse Person Re-Identification
- (CVPR 20) Relation-Aware Global Attention for Person Re-Identification
- (CVPR 14) DeepReID: Deep Filter Pairing Neural Network for Person Re-Identification
- (ICCV 15) Person Re-Identification with Correspondence Structure Learning
- (arXiv 17) AlignedReID: Surpassing Human-Level Performance in Person Re-Identification
- (CVPR 18) End-to-End Deep Kronecker-Product Matching for Person Re-Identification
- (ECCV 18) Beyond Part Models: Person Retrieval with Refined Part Pooling (and a Strong Convolutional Baseline)
- (ACM MM 18) Learning Discriminative Features with Multiple Granularities for Person Re-Identification
- (CVPR 19) Patch-based Discriminative Feature Learning for Unsupervised Person Re-Identification
- (CVPR 21) Diverse Part Discovery: Occluded Person Re-identification with Part-Aware Transformer
- (CVPR 17) Spindle Net: Person Re-Identification with Human Body Region Guided Feature Decomposition and Fusion
- (ICCV 17) Deeply-Learned Part-Aligned Representations for Person Re-Identification
- (ICCV 17) Pose-driven Deep Convolutional Model for Person Re-Identification
- (CVPR 18) Pose Transferrable Person Re-Identification
- (ECCV 18) Pose-Normalized Image Generation for Person Re-Identification
- (CVPR 18) Human Semantic Parsing for Person Re-Identification
- (NeurIPS 18) FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-Identification
- (TIP 19) Pose-Invariant Embedding for Deep Person Re-Identification
- (CVPR 19) Densely Semantically Aligned Person Re-Identification
- (ECCV 20) Identity-Guided Human Semantic Parsing for Person Re-Identification
- (ACM MM 21) Pose-guided Inter- and Intra-part Relational Transformer for Occluded Person Re-Identification
- (CVPR 22) Motion-Aware Transformer For Occluded Person Re-identification
- (TPAMI 22) Pose-Guided Representation Learning for Person Re-Identification
- (TPAMI 22) Multi-Task Learning With Coarse Priors for Robust Part-Aware Person Re-Identification
- (TNNLS 22) Parameter-Efficient Person Re-identification in the 3D Space
- (ECCV 16) Deep Attributes Driven Multi-Camera Person Re-Identification
- (arXiv 17) Improving Person Re-identification by Attribute and Identity Learning
- (CVPR 18) Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification
- (CVPR 19) AANet: Attribute Attention Network for Person Re-Identifications
- (CVPR 99) Bayesian Multi-Camera Surveillance
- (ICCV 03) Tracking across Multiple Cameras with Disjoint Views
- (CVPR 04) Bridging the Gaps between Cameras
- (CVPR 09) Multi-Camera Activity Correlation Analysis
- (ECCV 16) Human Re-Identification in Crowd Videos Using Personal, Social and Environmental Constraints
- (MMM 16) Camera Network Based Person Re-Identification by Leveraging Spatial-Temporal Constraint and Multiple Cameras Relations
- (arXiv 17) Distance-based Camera Network Topology Inference for Person Re-identification
- (ICCVW 17) Unified Framework for Automated Person Re-Identification and Camera Network Topology Inference in Camera Networks
- (CVPR 18) Unsupervised Cross-Dataset Person Re-Identification by Transfer Learning of Spatial-Temporal Patterns
- (AAAI 19) Spatial-Temporal Person Re-Identification
- (ICCV 21) Learning Instance-Level Spatial-Temporal Patterns for Person Re-Identification
- (ACM MM 21) MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification
Unsupervised domain adaptation (Rank-1 Rank-5 mAP)
| method | reference | DukeMTMC->Market-1501 | Market-1501->DukeMTMC |
|---|---|---|---|
| IDM | ICCV 21 | 93.2 97.5 82.8 | 83.6 91.5 93.7 |
| GLT | CVPR 21 | 92.2 96.5 79.5 | 82.0 90.2 69.2 |
| UNRN | AAAI 21 | 91.9 96.1 78.1 | 82.0 90.7 69.1 |
| SpCL | NeurIPS 20 | 90.3 96.2 76.7 | 82.9 90.1 68.8 |
| MEB-Net | ECCV 20 | 89.9 96.0 76.0 | 79.6 88.3 66.1 |
| MMT | ICLR 20 | 87.7 94.9 71.2 | 78.0 88.8 65.1 |
| CycAs | ECCV 20 | 84.8 --:- 64.8 | 77.9 --:- 60.1 |
| AD-Cluster | CVPR 20 | 86.7 94.4 68.3 | 72.6 82.5 54.1 |
| SNR | CVPR 20 | 82.8 --:- 61.7 | 76.3 --:- 58.1 |
| JVTC | ECCV 20 | 83.8 93.0 61.1 | 75.0 85.1 56.2 |
| ACT | AAAI 20 | 80.5 --:- 60.6 | 72.4 --:- 54.5 |
| MMCL | CVPR 20 | 84.4 92.8 60.4 | 72.4 82.9 51.3 |
| SSG | ICCV 19 | 80.0 90.0 58.3 | 73.0 80.6 53.4 |
| HCT | CVPR 20 | 80.0 91.6 56.4 | 69.6 83.4 50.7 |
| PAST | ICCV 19 | 78.4 --:- 54.6 | 72.4 ---:- 54.3 |
| PDA-Net | ICCV 19 | 75.2 86.3 47.6 | 63.2 77.0 45.1 |
| ECN | CVPR 19 | 75.1 87.6 43.0 | 63.3 75.8 40.4 |
| PAUL | CVPR 19 | 66.7 --:- 36.8 | 56.1 --:- 35.7 |
| CASCL | ICCV 19 | 64.7 80.2 35.6 | 51.5 66.7 30.5 |
| SPGAN | CVPR 18 | 58.1 76.0 26.9 | 46.9 62.6 26.4 |
| ATNet | ICCV 19 | 55.7 73.2 25.6 | 45.1 59.5 24.9 |
Other experiment settings: MSMT17->Market-1501, MSMT17->DukeMTMC, Market-1501->MSMT17 DukeMTMC->MSMT17, CUHK03->Market-1501, CUHK03->DukeMTMC, Market-1501->PRID2011
Unsupervised learning (Rank-1 mAP)
| method | reference | type | Market1-501 | DukeMTMC |
|---|---|---|---|---|
| PPLR | CVPR 22 | clustering | 94.3 84.4 | --:- --:- |
| ICE | ICCV 21 | contrastive | 93.8 82.3 | 83.3 69.9 |
| CAP | AAAI 21 | clustering | 91.4 79.2 | 81.1 67.3 |
| IICS | CVPR 21 | clustering | 89.5 72.9 | 80.0 64.4 |
| UGA | ICCV 19 | tracklet | 87.2 70.3 | 75.0 53.3 |
| MetaCam | CVPR 21 | clustering | 83.9 61.7 | 73.8 53.8 |
| TAUDL | ECCV 18 | tracklet | 63.7 41.2 | 61.7 43.5 |
| BUC | AAAI 19 | clustering | 66.2 38.3 | 47.4 27.5 |
Wei-Shi Zheng (SYSU)
| proc. | title | method | motivation |
|---|---|---|---|
| CVPR 20 | Weakly supervised discriminative feature learning with state information for person identification | WDBR + DFDR: classification loss with MPI-driven decision + state sub-distribution drift regularization | state-relevant feature distortion |
| ICCV 19 | Unsupervised Person Re-Identification by Camera-Aware Similarity Consistency Learning | CASCL: similarity consistency + intra-camera similarity preserving, global -> top-k neighbor | inconsistent similarity distribution (intra-camera similarity more reliable) |
| CVPR 19 | Patch-based Discriminative Feature Learning for Unsupervised Person Re-identification | PAUL: discriminative learning with patch feature bank, image-level triplet loss | the gap of similar patches is smaller than similar images |
| CVPR 19 | Unsupervised Person Re-identification by Soft Multilabel Learning | MAR: soft multilabel guided negative mining, similarity consistency, cross-view consistent learning | soft multilabel encodes relative comparative characteristic |
| CVPR 19 | Weakly Supervised Person Re-Identification | CV-MIML: intra-bag alignment, cross-view bag alignment | video-level weak label |
| ICCV 17 | Cross-view Asymmetric Metric Learning for Unsupervised Person Re-identification | CAMEL: project data into shared space, clustering | view-specific feature distortion |
Yi Yang (UTS)
| proc. | title | method | motivation |
|---|---|---|---|
| CVPR 19 | Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-identification | ECN: exemplar memory for invariance learning | intra-domain variations, exemplar-, camera-, neighborhood-invariance |
| AAAI 19 | A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification | BUC: bottom-up clustering + repelled loss training, diversity regularization | similarity and diversity in training data as supervision |
| ECCV 18 | Generalizing A Person Retrieval Model Hetero- and Homogeneously | HHL: camera invariance + domain correctness | intra- and inter- domain variance |
| CVPR 18 | Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning | EUG: select more pseudo-labeled tracklet for training during iteration, distance-based sampling criterion | initial pseudo-label predictions are unreliable |
| CVPR 18 | Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification | SPGAN: contrastive loss on self-similarity and domain-dissimilarity | ID should be preserved after image translation, domains contain entirely different ID sets |
| ICCV 17 | Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in vitro | LSRO: GAN, assign uniform label to generated data | to use unlabeled data |
Yonghong Tian (PKU)
| proc. | title | method | motivation |
|---|---|---|---|
| ECCV 20 | Multiple Expert Brainstorming for Domain Adaptive Person Re-identification | MEB-Net: MMT between 3 different models + ensemble based on scatter | ensemble learning remains unexplored |
| CVPR 20 | AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-identification | generate samples with diverse camera styles | low diversity in cluster |
Shiliang Zhang (PKU)
| proc. | title | method | motivation |
|---|---|---|---|
| ECCV 20 | Joint Visual and Temporal Consistency for Unsupervised Domain Adaptive Person Re-Identification | JVTC: local classification + global classification with temporal prob | a batch contains different id, temporal consistency |
| CVPR 20 | Unsupervised Person Re-identification via Multi-label Classification | MMCL: multi-label + memory |
Shaogang Gong (QMUL)
| proc. | title | method | motivation |
|---|---|---|---|
| ECCV 18 | Unsupervised Person Re-identification by Deep Learning Tracklet Association | TAUDL: dicriminate sparse tracklets within-camera, align similar cross-camera tracklets in mini-batch | source and target domains not always share some common characteristic |
| CVPR 18 | Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification | TJ-AIDL: joint learning + Identity Inferred Attribute + attribute consistency | heterogeneous attribute supervision |
Pong C Yuen (HKBU)
| proc. | title | method | motivation |
|---|---|---|---|
| ECCV 18 | Robust Anchor Embedding for Unsupervised Video Person Re-Identification in the Wild | RACE: embedding weight of kNN anchors | imbalanced unlabeled data, scalability |
| ICCV 17 | Dynamic Label Graph Matching for Unsupervised Video Re-Identification | DGM: cross-camera graph matching by neighborhood cost with re-weighed labels | cross-camera label estimation |
Others authors
AAAI 21
| from | title | method | motivation |
|---|---|---|---|
| USTC | Exploiting Sample Uncertainty for Domain Adaptive Person Re-Identification | UNRN: weighted contrastive loss | consistency between teacher & student |
| ZJU | Camera-aware Proxies for Unsupervised Person Re-Identification | CAP: intra-camera & inter-camera CL | camera variance |
ECCV 20 (3/5) (Yonghong Tian: 1, Shiliang Zhang: 1)
| from | title | method | motivation |
|---|---|---|---|
| UAE | Interpretable and Generalizable Person Re-identification with Query-adaptive Convolution and Temporal Lifting | QAConv: query-adaptive convolutional + TLift: |
generalizablity |
| THU | CycAs: Self-supervised Cycle Association for Learning Re-identifiable Descriptions | CycAs: cycle association on asymmetric pairs | data association in MOT |
| CMU | Joint Disentangling and Adaptation for Cross-Domain Person Re-Identification | DG-Net++: DG-Net + adversarial alignment + identification loss | disentangling + adaptation |
CVPR 20 (3/6) (Wei-shi Zheng: 1, Yonghong Tian: 1, Shiliang Zhang: 1)
| from | title | method | motivation |
|---|---|---|---|
| Huawei | Unsupervised Person Re-identification via Softened Similarity Learning | softened similarity + part + cross-camera encouragement | discard hard quantization loss |
| USTC | Style Normalization and Restitution for Generalizable Person Re-identification | SNR: style normalization + distill id-relevant feature | filter out id-irrelevant feature |
| NUDT | Hierarchical Clustering with Hard-batch Triplet Loss for Person Re-identification | HCT: hierarchinal clustering + hard-batch | pseudo label quality |
ICLR 20 (1/1)
| from | title | method | motivation |
|---|---|---|---|
| CUHK | Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification | MMT: optimize with soft pseudo label generated from the other network | noisy pseudo labels caused by clustering |
AAAI 20 (1/1)
| from | title | method | motivation |
|---|---|---|---|
| XMU | Asymmetric Co-Teaching for Unsupervised Cross-Domain Person Re-Identification | ACT: asymetric co-teaching trained by DBSCAN labeled inliners and outliers | noisy pseudo labels, consider outliers |
ICCV 19 (5/6) (Wei-shi Zheng: 1)
| from | title | method | motivation |
|---|---|---|---|
| NJU | A Novel Unsupervised Camera-aware Domain Adaptation Framework for Person Re-identification | CCE: cross-domain camera equiprobability, triplet selection in fragments | camera-level subdomains, temporal continuity |
| NTU | Cross-Dataset Person Re-Identification via Unsupervised Pose Disentanglement and Adaptation | PDA-Net: pose-guided image recovery and domain translation | jointly learn domain- and pose-invariant representation |
| UIUC | Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification | SSG: 3 sets of groups, triplet loss; semi-supervised | potential similarity from global to local |
| Tongji | Self-Training With Progressive Augmentation for Unsupervised Cross-Domain Person Re-Identification | PAST: 2-stage: CTL+RTL, global classification loss (HDBSCAN clustering, mean feature classifier) | unreliable pseudo labels, triplets focus on local info |
| CASIA | Unsupervised Graph Association for Person Re-identification | UGA: intra-camera tracklet classification, pull inter-camera pairs | underlying positive pair, lower GPU memory |
CVPR 19 (1/5) (Wei-shi Zheng: 3, Yi Yang: 1)
| from | title | method | motivation |
|---|---|---|---|
| USTC | Adaptive Transfer Network for Cross-Domain Person Re-Identification | ATNet: decompose domain transfer to sub-tasks, adaptive ensemble to each image | inter-domain disparities with multiple factors |
ECCV 18 (1/4)
| from | title | method | motivation |
|---|---|---|---|
| Argo AI | Domain Adaptation through Synthesis for Unsupervised Person Re-identification | SyRI: virtual human rendered with HDR maps | diversity in illumination |
CVPR 18 (2/5)
| from | title | method | motivation |
|---|---|---|---|
| SCUT | Unsupervised Cross-dataset Person Re-identification by Transfer Learning of Spatial-Temporal Patterns | TFusion: bayesian fusion, learn to rank | ST indecent of feature, constant FP, FN |
| TRACE | Disentangled Person Image Generation | disentangle FG, BG, pose | explicitly guide image genration |
ICCV 17 (4/7)
| from | title | method | motivation |
|---|---|---|---|
| NU | Efficient Online Local Metric Adaptation via Negative Samples for Person Re-Identification | ||
| UniFI | Group Re-Identification via Unsupervised Transfer of Sparse Features Encoding | ||
| UH | SHaPE: A Novel Graph Theoretic Algorithm for Making Consensus-based Decisions in Person Re-identification Systems | SHaPE: ranking -> shortest Hamiltonian path, ACS | aggregate multiple results |
| DUT | Stepwise Metric Promotion for Unsupervised Video Person Re-identification |
CVPR 17 (2/2)
| from | title | method | motivation |
|---|---|---|---|
| Disney Research | One-Shot Metric Learning for Person Re-identification | ||
| UCR | Unsupervised Adaptive Re-identification in Open World Dynamic Camera Networks |
- (CVPR 19) Generalizable Person Re-Identification by Domain-Invariant Mapping Network
- (CVPR 20) Style Normalization and Restitution for Generalizable Person Re-Identification
- (ECCV 20) Generalizing Person Re-Identification by Camera-Aware Invariance Learning and Cross-Domain Mixup
- (AAAI 21) Dual Distribution Alignment Network for Generalizable Person Re-Identification
- (CVPR 21) Meta Batch-Instance Normalization for Generalizable Person Re-Identification
- (CVPR 21) Generalizable Person Re-Identification with Relevance-aware Mixture of Experts
- (AVSS 19) Continuous Learning without Forgetting for Person Re-Identification
- (AAAI 21) Generalising without Forgetting for Lifelong Person Re-Identification
- (WACV 21) Continual Representation Learning for Biometric Identification
- (CVPR 21) Lifelong Person Re-Identification via Adaptive Knowledge Accumulation
- (AAAI 22) Lifelong Person Re-identification by Pseudo Task Knowledge Preservation
- (CVPR 22) Lifelong Unsupervised Domain Adaptive Person Re-Identification with Coordinated Anti-forgetting and Adaptation
- (ACM MM 22) Meta Reconciliation Normalization for Lifelong Person Re-Identification
- (ACM MM 22) Patch-based Knowledge Distillation for Lifelong Person Re-Identification
- (ICCV 17) RGB-Infrared Cross-Modality Person Re-Identification
- (IJCAI 18) Cross-Modality Person Re-Identification with Generative Adversarial Training
- (CVPR 19) Learning to Reduce Dual-Level Discrepancy for Infrared-Visible Person Re-Identification
- (ICCV 19) RGB-Infrared Cross-Modality Person Re-Identification via Joint Pixel and Feature Alignment
- (AAAI 20) Cross-Modality Paired-Images Generation for RGB-Infrared Person Re-Identification
- (CVPR 20) Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification
- (CVPR 20) Cross-Modality Person Re-Identification with Shared-Specific Feature Transfer
- (ECCV 20) Dynamic Dual-Attentive Aggregation Learning for Visible-Infrared Person Re-Identification
- (CVPR 21) Neural Feature Search for RGB-Infrared Person Re-Identification
- (CVPR 21) Farewell to Mutual Information: Variational Distillation for Cross-Modal Person Re-Identification
- (CVPR 21) Discover Cross-Modality Nuances for Visible-Infrared Person Re-Identification
- (ICCV 21) Syncretic Modality Collaborative Learning for Visible Infrared Person Re-Identification
- (ICCV 21) Learning by Aligning: Visible-Infrared Person Re-Identification Using Cross-Modal Correspondences
- (ICCV 21) Cross-Modality Person Re-Identification via Modality Confusion and Center Aggregation
- SYSU-MM01: MCLNet
- RegDB: SMCL
- (TPAMI 19) Person Re-Identification by Contour Sketch under Moderate Clothing Change
- (IJCNN 19) Celebrities-ReID: A Benchmark for Clothes Variation in Long-Term Person Re-Identification
- (ACCV 20) Long-Term Cloth-Changing Person Re-Identification
- (CVPR 20) COCAS: A Large-Scale Clothes Changing Person Dataset for Re-Identification
- (BMVC 09) Associating Groups of People
- (ICCV 17) Group Re-Identification via Unsupervised Transfer of Sparse Features Encoding
- (ACM MM 18) Group Re-Identification: Leveraging and Integrating Multi-Grain Information
- (CVPR 19) Learning Context Graph for Person Search
- (CVPR 22) Modeling 3D Layout for Group Re-Identification
Duke: DukeMTMC group dataset, Road: Road group dataset
| from | method | motivation | Duke rank1 | Duke rank5 | Road rank1 | Road rank5 |
|---|---|---|---|---|---|---|
| QMUL | (BMVC 09) CRRRO-BRO: 2 descriptors of ring regions | rotation invariant | 9.9 | 26.1 | 17.8 | 34.6 |
| OYO | (ICPR 10) covariance descriptor | illumination invariance | 21.3 | 43.6 | 38.9 | 61.0 |
| USTC | (ICIP 16) BSC+CM: salience channel +consistent matching | unreliable patch matches | 23.1 | 44.3 | 58.6 | 80.6 |
| UniFI | (ICCV 17) PREF: mask + sparse encoding | spatial displacement | 22.3 | 44.3 | 43.0 | 68.7 |
| SJTU | (MM 18) MGR: multi-grain representation + matching | group granularity | 47.4 | 68.1 | 72.3 | 90.6 |
| SJTU | (TCYB 19) hand=>conv | utilize dl | 48.4 | 75.2 | 80.2 | 93.8 |
| NTHU | (arXiv 19) DotSCN: learn single&couple representation on transferred reID dataset | too few images to learn group representation | 86.4 | 98.8 | 84.0 | 95.1 |
- (ACM MM 14) Person Search in a Scene by Jointly Modeling People Commonness and Person Uniqueness
- (CVPR 17) Joint Detection and Identification Feature Learning for Person Search
- (ICCV 17) Neural Person Search Machines
- (ICCV 19) Re-ID Driven Localization Refinement for Person Search
- (CVPR 20) Robust Partial Matching for Person Search in the Wild
- (CVPR 20) Norm-Aware Embedding for Efficient Person Search
- (CVPR 21) Anchor-Free Person Search
| from | method | motivation | mAP (CUHK-SYSU) | top-1 (CUHK-SYSU) | mAP (PRW) | top-1 (PRW) |
|---|---|---|---|---|---|---|
| CUHK | (CVPR 17) OIM: online matching with lookup table | joint detection and reID | 75.5 | 78.7 | 21.3 | 49.9 |
| TUM | => larger image | 83.3 | 84.2 | 38.3 | 70.0 | |
| XJTU | (arXiv 2017) IAN: center loss | intra-class variance | 77.2 | 80.5 | 23.0 | 61.9 |
| HFUT | (ICCV 17) NPSM: recursive region shrinking with lstm | query-guided proposal | 77.9 | 81.2 | 24.2 | 53.0 |
| CMU | (ECCV 18) RCAA: context-aware agent modifies bbox | query-guided proposal with RL | 79.5 | 81.3 | - | - |
| NJUST | (ECCV 18) MGTS: mask-guided two-stream | emphasize foreground info | 83.0 | 83.7 | 32.6 | 72.1 |
| QMUL | (ECCV 18) CLSA: in-network feature pyramid | multi-scale feature | 87.2 | 88.5 | 38.7 | 65.0 |
| SJTU | (CVPR 19) contextual instance graph | learn contexual info | 84.1 | 86.5 | 33.4 | 73.6 |
| TUM | (CVPR 19) QEEPS: query-guided SSE-Net, RPN, simNet | leverage query extensively | 88.9 | 89.1 | 39.1 | 80.0 |
| HUST | (ICCV 19) regression => reID gradient | refine detection | 93.0 | 94.2 | 42.9 | 70.2 |
person search = detection + re-identification
| Dataset | from | time | description | # identities | # cameras | # images |
|---|---|---|---|---|---|---|
| VIPeR | UCSC | 2007 | - | 632 | 2 | 1264 |
| QMUL iLIDS | QMUL | 2009 | airport | 119 | 2 | 476 |
| GRID | QMUL | 2009 | underground | 1025 | 8 | 1275 |
| PRID2011 | TUGraz | 2011 | campus | 934 | 2 | 24541 |
| CUHK01 | CUHK | 2012 | campus | 971 | 2 | 3884 |
| CUHK02 | CUHK | 2013 | campus | 1816 | 10(5 pairs) | 7264 |
| CUHK03 | CUHK | 2014 | campus | 1467 | 10(5 pairs) | 13164 |
| iLIDS-VID | QMUL | 2014 | iLIDS video | 300 | 2 | 42495 |
| Market1501 | THU | 2015 | campus supermarket | 1501 | 6 | 32217 |
| PKU-Reid | PKU | 2016 | 8 orientations | 114 | 2 | 1824 |
| PRW | UTS | 2016 | Market1501 full frame | 932 | 6 | 34304 |
| MARS | THU | 2016 | Market1501 video | 1261 | 6 | 1191003 |
| DukeMTMC-reID | Duke | 2017 | campus | 1812 | 8 | 36441 |
| DukeMTMC4RID | Duke | 2017 | Duke SOTA detector | 1852 | 8 | 46261 |
| MSMT17 | PKU | 2018 | scene & lighting change | 4101 | 15 | 126441 |