Person Re-identification on CUHK03 NP (new protocol) (test)
76.7mAPR-Erasing
Evaluation results
| Method | Links | ||
|---|---|---|---|
| R-ErasingTypes=Real Aug.2026.04 | 76.7 | 78.4 | |
| reinforcement-guided synthetic data generation frameworkTypes=Synthetic Aug.2026.04 | 76.6 | 79.3 | |
| ResNet-50Types=Base2026.04 | 74.1 | 76.5 | |
| IDiffTypes=Synthetic Aug.2026.04 | 73.1 | 75.4 | |
| GIF-SDTypes=Synthetic Aug.2026.04 | 71.7 | 74.6 | |
| TriNet + Random Erasing + re-rankingBackbone=ResNet-50, Data Augmentation=Random Erasing, Re-ranking=true2017.08 | 65.05 | 63.93 | |
| DG-Net2019.04 | 61.1 | 65.6 | |
| PCB (RPP)Query mode=single-query, Pooling=refined part pooling2017.11 | 57.5 | 63.7 | |
| PCBpooling=Refined Part Pooling (RPP)2017.11 | 57.5 | 63.7 | |
| PCB + RPP2018.04 | 57.5 | 63.7 | |
| HPMpyramid scales=4, pooling=average and max pooling2018.04 | 57.5 | 63.9 | |
| PCB + RPP2019.04 | 57.5 | 63.7 | |
| CaAugTypes=Real Aug.2026.04 | 57.4 | 59.4 | |
| PCB (UP)Query mode=single-query, Pooling=uniform partition2017.11 | 54.2 | 61.3 | |
| PCBpooling=Uniform Partition (UP)2017.11 | 54.2 | 61.3 | |
| PCB2018.04 | 54.2 | 61.3 | |
| PCB2019.04 | 54.2 | 61.3 | |
| TriNet + Random ErasingBackbone=ResNet-50, Data Augmentation=Random Erasing2017.08 | 53.83 | 58.14 | |
| TriNet+EraQuery mode=single-query, Data Augmentation=Random Erasing2017.11 | 50.7 | 55.5 | |
| TriNet+EraData Augmentation=Era2017.11 | 50.7 | 55.5 | |
| MLFN2018.04 | 49.2 | 54.7 | |
| MLFN2019.04 | 47.8 | 52.8 | |
| TriNetBackbone=ResNet-502017.08 | 46.74 | 49.86 | |
| OG-Net-DeepInput Type=point clouds, Loss Function=CE + Circle, #params(M)=2.472020.06 | 45.71 | 49.43 | |
| OG-NetInput Type=point clouds, Loss Function=CE + Circle, #params(M)=1.952020.06 | 43.73 | 48.29 | |
| SVDNet+EraQuery mode=single-query, Data Augmentation=Random Erasing2017.11 | 43.5 | 48.7 | |
| SVDNet+EraData Augmentation=Era2017.11 | 43.5 | 48.7 | |
| OG-Net-SmallInput Type=point clouds, Loss Function=CE + Circle, #params(M)=1.202020.06 | 41.79 | 46.43 | |
| OG-Net-DeepInput Type=point clouds, Loss Function=CE, #params(M)=2.472020.06 | 41.15 | 45.71 | |
| DPFLBackbone=ResNet-502017.08 | 40.5 | 43 | |
| OG-NetInput Type=point clouds, Loss Function=CE, #params(M)=1.952020.06 | 39.28 | 44 | |
| PT2019.04 | 38.7 | 41.6 | |
| HA-CNN2018.04 | 38.6 | 41.7 | |
| HA-CNN2019.04 | 38.6 | 41.7 | |
| OG-Net-SmallInput Type=point clouds, Loss Function=CE, #params(M)=1.202020.06 | 38.06 | 43.07 | |
| SVDNetBackbone=ResNet-502017.08 | 37.8 | 40.9 | |
| SVDNetBackbone=ResNet-50, Post-processing=None2017.03 | 37.3 | 41.5 | |
| SVDNetQuery mode=single-query2017.11 | 37.3 | 41.5 | |
| SVDNet2017.11 | 37.3 | 41.5 | |
| SVDNet2018.04 | 37.3 | 41.5 | |
| MultiScaleQuery mode=single-query2017.11 | 37 | 40.7 | |
| MultiScale2017.11 | 37 | 40.7 | |
| MultiScale2018.04 | 37 | 40.7 | |
| FineGPRTypes=Simulated Aug.2026.04 | 36.4 | 37.9 | |
| ResNet-50Input Type=images, Loss Function=CE + Circle, #params(M)=24.562020.06 | 34.12 | 37.29 | |
| PANBackbone=ResNet-50, Post-processing=None2017.03 | 34 | 36.3 | |
| PANQuery mode=single-query2017.11 | 34 | 36.3 | |
| PAN2017.11 | 34 | 36.3 | |
| PAN2018.04 | 34 | 36.3 | |
| DenseNet-121Input Type=images, Loss Function=CE + Circle, #params(M)=8.502020.06 | 33.52 | 36.21 | |
| M³LSource=MS+D+M, IDs=2,494, Images=62,079, Backbone=IBN-Net502020.12 | 32.1 | 33.1 | |
| M³LVenue=CVPR'21, Training=Multi2021.04 | 32.1 | 33.1 | |
| ResNet-50Input Type=images, Loss Function=CE, #params(M)=24.562020.06 | 32.09 | 35.43 | |
| IDE+DaFBackbone=ResNet-502017.08 | 31.5 | 27.5 | |
| M³LSource=MS+D+M+C-NP, IDs=2,494, Images=62,079, Backbone=IBN-Net50, evaluation_subset=detected2020.12 | 31.4 | 31.6 | |
| M³LSource=MS+D+M+C-NP, IDs=2,494, Images=62,079, Backbone=ResNet-50, evaluation_subset=detected2020.12 | 30.9 | 31.9 | |
| M³LSource=MS+D+M, IDs=2,494, Images=62,079, Backbone=ResNet-502020.12 | 29.9 | 30.7 | |
| DenseNet-121Input Type=images, Loss Function=CE, #params(M)=8.502020.06 | 29.45 | 33.64 | |
| QAConv-GSVenue=Ours, Training=MSMT17 (all)2021.04 | 28 | 27.6 | |
| MobileNetV2Input Type=images, Loss Function=CE, #params(M)=4.162020.06 | 26.45 | 29.57 | |
| MobileNetV2Input Type=images, Loss Function=CE + Circle, #params(M)=4.162020.06 | 25.46 | 29.14 | |
| SVDNetBackbone=CaffeNet, Post-processing=None2017.03 | 24.9 | 27.7 | |
| InfinitePersonTypes=Simulated Aug.2026.04 | 24.7 | 24.6 | |
| TransMatcherTraining Dataset=ClonedPerson (CP), #ID=4,826, #Imgs=763k2022.04 | 24.4 | — | |
| ShuffleNetV2Input Type=images, Loss Function=CE + Circle, #params(M)=1.782020.06 | 23.56 | 25.43 | |
| TransMatcherTraining Dataset=RandPerson* (RP*), #ID=8,000, #Imgs=1,239k2022.04 | 22.9 | — | |
| ShuffleNetV2Input Type=images, Loss Function=CE, #params(M)=1.782020.06 | 22.9 | 25.29 | |
| QAConv50Source=Com-MS, IDs=4,101, Images=126,4412020.12 | 22.6 | 25.3 | |
| QAConvVenue=ECCV'20, Training=MSMT17 (all)2021.04 | 22.6 | 25.3 | |
| QAConvTraining Dataset=ClonedPerson (CP), #ID=4,826, #Imgs=763k2022.04 | 21.8 | — | |
| IDEBackbone=ResNet-502017.08 | 21 | 22.2 | |
| QAConv50Source=MS+D+M, IDs=2,494, Images=62,079, reimplemented=true2020.12 | 21 | 23.5 | |
| QAConv-GSVenue=Ours, Training=MSMT172021.04 | 20.6 | 20.9 | |
| QAConvTraining Dataset=RandPerson* (RP*), #ID=8,000, #Imgs=1,239k2022.04 | 20.1 | — | |
| PointNet++ (MSG)Input Type=point clouds, Loss Function=CE + Circle, #params(M)=1.872020.06 | 19.86 | 21.36 | |
| PointNet++ (MSG)Input Type=point clouds, Loss Function=CE, #params(M)=1.872020.06 | 19.79 | 21.14 | |
| BaselineBackbone=ResNet-50, Post-processing=None2017.03 | 19.7 | 21.3 | |
| TransMatcherTraining Dataset=UnrealPerson (UP), #ID=6,799, #Imgs=1,256k2022.04 | 19.7 | — | |
| TransMatcherTraining Dataset=UnrealPerson (UP), #ID=3,000, #Imgs=120k2022.04 | 19.6 | — | |
| QAConv50Source=MS+D+M+C-NP, IDs=2,494, Images=62,079, reimplemented=true, evaluation_subset=detected2020.12 | 19.2 | 22.9 | |
| TransMatcherTraining Dataset=RandPerson (RP), #ID=8,000, #Imgs=1,801k2022.04 | 18.7 | — | |
| QAConv-GSVenue=Ours, Training=Market-15012021.04 | 18.1 | 19.1 | |
| QAConvTraining Dataset=UnrealPerson (UP), #ID=3,000, #Imgs=120k2022.04 | 17.8 | — | |
| QAConvTraining Dataset=UnrealPerson (UP), #ID=6,799, #Imgs=1,256k2022.04 | 17.2 | — | |
| TransMatcherTraining Dataset=RandPerson (RP), #ID=8,000, #Imgs=132k2022.04 | 16.9 | — | |
| QAConv-GSVenue=Ours, Training=RandPerson2021.04 | 16.1 | 18.4 | |
| QAConvTraining Dataset=RandPerson (RP), #ID=8,000, #Imgs=1,801k2022.04 | 16 | — | |
| QAConvTraining Dataset=RandPerson (RP), #ID=8,000, #Imgs=132k2022.04 | 15.1 | — | |
| PointNet++ (SSG)Input Type=point clouds, Loss Function=CE, #params(M)=1.592020.06 | 13.97 | 14.57 | |
| LOMO+XQDABackbone=ResNet-502017.08 | 13.6 | 14.8 | |
| SpCLTraining Dataset=ClonedPerson (CP), #ID=4,826, #Imgs=75k2022.04 | 12 | — | |
| LOMO+XQDAPost-processing=None2017.03 | 11.5 | 12.8 | |
| LOMO+XQDAQuery mode=single-query2017.11 | 11.5 | 12.8 | |
| LOMO+XQDA2017.11 | 11.5 | 12.8 | |
| RP BaselineVenue=ACMMM'20, Training=RandPerson2021.04 | 10.8 | 13.4 | |
| MuDeepVenue=TPAMI'20, Training=Market-15012021.04 | 9.1 | 10.3 | |
| QAConvVenue=ECCV'20, Training=Market-15012021.04 | 8.6 | 9.9 | |
| MGNVenue=ACMMM'18, Training=Market-15012021.04 | 7.4 | 8.5 | |
| BOW+XQDABackbone=ResNet-502017.08 | 7.29 | 7.93 | |
| BoW+kissmePost-processing=None2017.03 | 6.4 | 6.4 |