Human Pose Estimation on COCO (test-dev)
81.1APViTPose
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ViTPoseBackbone=VITAE-G, Ensemble=true2022.04 | 81.1 | 95 | 88.2 | 77.8 | 86 | 85.6 | — | — | — | — | — | — | — | — | — | — | |
| ViTPoseBackbone=VITAE-G, Ensemble=false2022.04 | 80.9 | 94.8 | 88.1 | 77.5 | 85.9 | 85.4 | — | — | — | — | — | — | — | — | — | — | |
| UDP+++Backbone=HRNet-w48plus, Ensemble=true2022.04 | 80.8 | 94.9 | 88.1 | 77.4 | 85.7 | 85.3 | — | — | — | — | — | — | — | — | — | — | |
| SCAIBackbone=HR48, Size=384×2882023.03 | 80.6 | 94.8 | 87 | 78.1 | 84.8 | 83.1 | — | — | — | — | — | — | — | — | — | — | |
| UDP-Pose-PSA(p)Backbone=HRNet-W48, Input Size=384 x 288, Flops=35.4G, mPara=70.1M2021.07 | 79.5 | 93.6 | 85.9 | 76.3 | 84.3 | 81.9 | — | — | — | — | — | — | — | — | — | — | |
| UDP-Pose-PSA(s)Backbone=HRNet-W48, Input Size=384 x 288, Flops=35.4G, mPara=69.1M2021.07 | 79.4 | 93.6 | 85.8 | 76.1 | 84.1 | 81.7 | — | — | — | — | — | — | — | — | — | — | |
| SCIOBackbone=HR48, Size=384×2882022.07 | 79.2 | 93.5 | 85.8 | 74.1 | 84.2 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| RSN++Backbone=4xRSN-50, Ensemble=false2022.04 | 79.2 | 94.4 | 87.1 | 76.1 | 83.8 | 84.1 | — | — | — | — | — | — | — | — | — | — | |
| SCIOBackbone=HR48, Size=384×2882023.03 | 79.2 | 93.5 | 85.8 | 74.1 | 84.2 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDP-Pose-PSA(p)Backbone=HRNet-W48, Input Size=256 x 192, Flops=15.7G, mPara=70.1M2021.07 | 78.9 | 93.6 | 85.8 | 76.1 | 83.6 | 81.4 | — | — | — | — | — | — | — | — | — | — | |
| CCMBackbone=HRNet-w48, Ensemble=true2022.04 | 78.9 | 93.8 | 86 | 75 | 84.5 | 83.6 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1*+AIDBackbone=HRNet-W48plus, Input size=384 x 288, Paradigm=Top-down, Extra data=true2020.08 | 78.7 | 93.1 | 85.9 | 75.2 | 84.5 | 84 | — | — | — | — | — | — | — | — | — | — | |
| PoseBHBackbone=Huge2025.05 | 78.6 | 93.3 | 86.2 | 75.3 | 84.4 | 83.5 | — | — | — | — | — | — | — | — | — | — | |
| ViTPose++Backbone=Huge2025.05 | 78.5 | 93.4 | 86.2 | 75.3 | 84.4 | 83.4 | — | — | — | — | — | — | — | — | — | — | |
| PoseurBackbone=HRFormer-B, Paradigm=Regression-based, Number of Decoders=6, Input Resolution=384 x 2882022.01 | 78.3 | 93.5 | 85.9 | 75.2 | 83.4 | — | — | — | — | — | — | — | — | — | — | — | |
| UDPv1*Backbone=HRNet-W48plus, Input size=384 x 288, Paradigm=Top-down, Extra data=true2020.08 | 78.2 | 92.8 | 85.4 | 74.8 | 84.2 | 83.6 | — | — | — | — | — | — | — | — | — | — | |
| MegviiApproach Type=Top-Down2018.12 | 78.1 | 94.1 | 85.9 | 74.5 | 83.3 | — | — | — | — | — | — | — | — | — | — | — | |
| MSPN++Backbone=4xResNet-50, Ensemble=false2022.04 | 78.1 | 94.1 | 85.9 | 74.5 | 83.3 | 83.1 | — | — | — | — | — | — | — | — | — | — | |
| UDP-Pose (Strong Baseline)Backbone=HRNet-W48, Input Size=384 x 288, Flops=33.0G, mPara=63.8M, Extra training data=Yes2021.07 | 77.8 | 92 | 84.3 | 74.2 | 84.5 | 82.5 | — | — | — | — | — | — | — | — | — | — | |
| PoseurBackbone=HRNet-W48, Paradigm=Regression-based, Number of Decoders=6, Input Resolution=384 x 2882022.01 | 77.6 | 92.9 | 85 | 74.4 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | |
| UDPv1+AIDBackbone=HRNet-W48plus, Input size=384 x 288, Paradigm=Top-down2020.08 | 77.5 | 92.9 | 84.9 | 74.2 | 83.4 | 83 | — | — | — | — | — | — | — | — | — | — | |
| Ours (+DARK)Network=HRNetW48, Input Size=384 x 2882023.03 | 77.5 | — | — | — | — | 82.4 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPs=32.9, Extra data=true2019.10 | 77.4 | 92.6 | 84.6 | 73.6 | 83.7 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| DARK*Backbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down, Extra data=true2020.08 | 77.4 | 92.6 | 84.6 | 73.6 | 83.7 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HRNet-w48, Ensemble=false2022.04 | 77.4 | 92.6 | 84.6 | 73.6 | 83.7 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| UDP-PoseBackbone=HRNet-W48, Input Size=256 x 192, Flops=14.7G, mPara=63.8M2021.07 | 77.2 | 91.8 | 83.7 | 73.8 | 83.7 | 82 | — | — | — | — | — | — | — | — | — | — | |
| DUAL (+DARK)Network=HRNetW48, Input Size=384 x 2882023.03 | 77.2 | — | — | — | — | 82.2 | — | — | — | — | — | — | — | — | — | — | |
| MSPNBackbone=MSPN, Input size=384 x 288, Paradigm=Top-down, Extra data=true2020.08 | 77.1 | 93.8 | 84.6 | 73.4 | 82.3 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48Backbone=HRNet-W48, Input size=384 x 288, #Params=63.6M, GFLOPS=32.9, Extra data=extra data, Paradigm=Top-down2019.02 | 77 | 92.7 | 84.5 | 73.4 | 83.1 | 82 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPs=32.9, Extra data=true2019.10 | 77 | 92.7 | 84.5 | 73.4 | 83.1 | 82 | — | — | — | — | — | — | — | — | — | — | |
| HRNet*Backbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down, Extra data=true2020.08 | 77 | 92.7 | 84.5 | 73.4 | 83.1 | 82 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1+AIDBackbone=HRNet-W32, Input size=384 x 288, Paradigm=Top-down2020.08 | 77 | 92.9 | 84.5 | 73.6 | 82.8 | 82.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-w48, Ensemble=false2022.04 | 77 | 92.7 | 84.5 | 73.4 | 83.1 | 82 | — | — | — | — | — | — | — | — | — | — | |
| Graph-PCNNBackbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down2020.08 | 76.8 | 92.6 | 84.3 | 73.3 | 82.7 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1Backbone=HRNet-W48plus, Input size=384 x 288, Paradigm=Top-down2020.08 | 76.8 | 92.8 | 84.1 | 73.4 | 82.6 | 82.4 | — | — | — | — | — | — | — | — | — | — | |
| Dark-PoseBackbone=HRNet-W48, Input Size=384 x 288, Flops=32.9G, mPara=63.6M2021.07 | 76.8 | 90.6 | 83.2 | 72.8 | 84 | 81.7 | — | — | — | — | — | — | — | — | — | — | |
| Graph-PCNNBackbone=HR48, Size=384×2882022.07 | 76.8 | 92.6 | 84.3 | 73.3 | 82.7 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| Graph-PCNNBackbone=HR48, Size=384×2882023.03 | 76.8 | 92.6 | 84.3 | 73.3 | 82.7 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| PoseFixBackbone=HRNet-W48+ResNet-152, Input size=384 x 288, Paradigm=Top-down2020.08 | 76.7 | 92.6 | 84.1 | 73.1 | 82.6 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1+AIDBackbone=HRNet-W48, Input size=256 x 192, Paradigm=Top-down2020.08 | 76.7 | 92.9 | 84.2 | 73.4 | 82.3 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| PoseFixBackbone=HR48+R152, Size=384×2882022.07 | 76.7 | 92.6 | 84.1 | 73.1 | 82.6 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| PoseFixBackbone=HR48+R152, Size=384×2882023.03 | 76.7 | 92.6 | 84.1 | 73.1 | 82.6 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1+AIDBackbone=2xRSN-50, Input size=256 x 192, Paradigm=Top-down2020.08 | 76.6 | 92.8 | 84.3 | 73.6 | 82.1 | 82.5 | — | — | — | — | — | — | — | — | — | — | |
| PoseBHBackbone=Base2025.05 | 76.6 | 92.6 | 84.4 | 73.4 | 82.4 | 81.7 | — | — | — | — | — | — | — | — | — | — | |
| MRSAApproach Type=Top-Down2018.12 | 76.5 | 92.4 | 84 | 73 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | |
| UDPv1Backbone=HRNet-W32, Input size=384 x 288, Paradigm=Top-down2020.08 | 76.5 | 92.6 | 83.8 | 72.9 | 82.5 | 82.1 | — | — | — | — | — | — | — | — | — | — | |
| HRNet + UDPBackbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down2019.11 | 76.5 | 92.7 | 84 | 73 | 82.4 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDPBackbone=HR48, Size=384×2882022.07 | 76.5 | 92.7 | 84 | 73 | 82.4 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDP-PoseBackbone=HRNet-W48, Paradigm=Heatmap-based, Input Resolution=384 x 2882022.01 | 76.5 | 92.7 | 84 | 73 | 82.4 | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet-152, Ensemble=true2022.04 | 76.5 | 92.4 | 84 | 73 | 82.7 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| UDPNetwork=HRNetW48, Input Size=384 x 2882023.03 | 76.5 | — | — | — | — | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDPBackbone=HR48, Size=384×2882023.03 | 76.5 | 92.7 | 84 | 73 | 82.4 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| ViTPose++Backbone=Base2025.05 | 76.4 | 92.7 | 84.3 | 73.2 | 82.2 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48Backbone=HRNet, Input Size=384 x 288, Flops=32.9G, mPara=63.6M2021.07 | 76.3 | 90.8 | 82.9 | 72.3 | 83.4 | 81.2 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPs=32.92019.10 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down2020.08 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1+AIDBackbone=HRNet-W32, Input size=256 x 192, Paradigm=Top-down2020.08 | 76.2 | 92.8 | 83.8 | 72.9 | 81.8 | 81.8 | — | — | — | — | — | — | — | — | — | — | |
| DARKbackbone=HRNetV1-W48, input size=384 x 288, #Params=63.6M, GFLOPS=32.92021.04 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| DARKPrediction Type=Heatmap based, Backbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPS=32.9, Multi-scale test=false2021.04 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| UDP-PoseBackbone=HRNet-W48, Input Size=384 x 288, Flops=33.0G, mPara=63.8M2021.07 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| HRFormer-Binput size=384 x 288, #param.=43.2M, FLOPS=26.8G, Pre-training=ImageNet2021.10 | 76.2 | 92.7 | 83.8 | 72.5 | 82.3 | 81.2 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HR48, Size=384×2882022.07 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| HRFormerBackbone=HRFormer-B, Paradigm=Heatmap-based, Input Resolution=384 x 2882022.01 | 76.2 | 92.7 | 83.8 | 72.5 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | |
| DARKNetwork=HRNetW48, Input Size=384 x 2882023.03 | 76.2 | — | — | — | — | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| DARKBackbone=HR48, Size=384×2882023.03 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1Backbone=HRNet-W48, Input size=256 x 192, Paradigm=Top-down2020.08 | 76.1 | 92.4 | 83.7 | 72.7 | 81.9 | 81.8 | — | — | — | — | — | — | — | — | — | — | |
| MSPNBackbone=MSPN, Input size=384 x 288, Paradigm=Top-down2019.11 | 76.1 | 93.4 | 83.8 | 72.3 | 81.5 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| HRNet + UDPBackbone=HRNet-W32, Input size=384 x 288, Paradigm=Top-down2019.11 | 76.1 | 92.5 | 83.5 | 72.8 | 82 | 81.3 | — | — | — | — | — | — | — | — | — | — | |
| MSPNBackbone=MSPN, Size=384×2882022.07 | 76.1 | 93.4 | 83.8 | 72.3 | 81.5 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| MSPNNetwork=ResNet50, Input Size=384 x 2882023.03 | 76.1 | — | — | — | — | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| MSPNBackbone=MSPN, Size=384×2882023.03 | 76.1 | 93.4 | 83.8 | 72.3 | 81.5 | 81.6 | — | — | — | — | — | — | — | — | — | — | |
| UDPv1Backbone=2xRSN-50, Input size=256 x 192, Paradigm=Top-down2020.08 | 76 | 92.5 | 83.6 | 73.2 | 81.4 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| The Sea MonstersApproach Type=Top-Down2018.12 | 75.9 | 92.1 | 83 | 71.7 | 82.1 | — | — | — | — | — | — | — | — | — | — | — | |
| TokenPose-L/D24input size=384 x 288, #param.=29.8M, FLOPS=22.1G, Pre-training=ImageNet2021.10 | 75.9 | 92.3 | 83.4 | 72.2 | 82.1 | 80.8 | — | — | — | — | — | — | — | — | — | — | |
| TokenPoseBackbone=L/D24, Paradigm=Heatmap-based, Input Resolution=384 x 2882022.01 | 75.9 | 92.3 | 83.4 | 72.2 | 82.1 | — | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W32Backbone=HRNet, Input Size=384 x 288, Flops=16.0G, mPara=28.5M2021.07 | 75.8 | 90.6 | 82.7 | 71.9 | 82.8 | 81 | — | — | — | — | — | — | — | — | — | — | |
| HRNet + UDPBackbone=HRNet-W48, Input size=256 x 192, Paradigm=Top-down2019.11 | 75.7 | 92.4 | 83.3 | 72.5 | 81.4 | 80.9 | — | — | — | — | — | — | — | — | — | — | |
| EvoPose2D-LBackbone=ResNet-1012021.07 | 75.7 | 91.9 | 83.1 | 72.2 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | |
| HRNet + RLEBackbone=HRNet-W482021.07 | 75.7 | 92.3 | 82.9 | 72.3 | 81.3 | — | — | — | — | — | — | — | — | — | — | — | |
| EvoPose2D-LStage=Two-stage, Test-Time Augmentation (TTA)=true2021.11 | 75.7 | 91.9 | 83.1 | 72.2 | 81.5 | 81.7 | — | — | — | — | — | — | — | — | — | — | |
| MIPNetStage=Two-stage, Test-Time Augmentation (TTA)=true2021.11 | 75.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RLEStage=Two-stage, Test-Time Augmentation (TTA)=true2021.11 | 75.7 | 92.3 | 82.9 | 72.3 | 81.3 | — | — | — | — | — | — | — | — | — | — | — | |
| RLEBackbone=HRNet-W48, Paradigm=Regression-based, Input Resolution=384 x 2882022.01 | 75.7 | 92.3 | 82.9 | 72.3 | 81.3 | — | — | — | — | — | — | — | — | — | — | — | |
| UDPv1Backbone=HRNet-W32, Input size=256 x 192, Paradigm=Top-down2020.08 | 75.6 | 92.3 | 83.2 | 72.3 | 81.4 | 81.4 | — | — | — | — | — | — | — | — | — | — | |
| Dark-PoseBackbone=HRNet-W32, Input Size=256 x 192, Flops=7.1G, mPara=28.5M2021.07 | 75.6 | 90.5 | 82.1 | 71.8 | 82.8 | 80.8 | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48Backbone=HRNet-W48, Input size=384 x 288, #Params=63.6M, GFLOPS=32.9, Paradigm=Top-down2019.02 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPs=32.92019.10 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down2020.08 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| RSNBackbone=2xRSN-50, Input size=256 x 192, Paradigm=Top-down2020.08 | 75.5 | 93.6 | 84 | 73 | 79.6 | 81.3 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Input size=384 x 288, Paradigm=Top-down2019.11 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetV1backbone=HRNetV1-W48, input size=384 x 288, #Params=63.6M, GFLOPS=32.92021.04 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetPrediction Type=Heatmap based, Backbone=HRNet-W48, Input size=384 × 288, #Params=63.6M, GFLOPS=32.9, Multi-scale test=false2021.04 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W482021.07 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48input size=384 x 288, #param.=63.6M, FLOPS=32.9G, Pre-training=ImageNet2021.10 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HR48, Size=384×2882022.07 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48Stage=Two-stage, Test-Time Augmentation (TTA)=true2021.11 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Paradigm=Heatmap-based, Input Resolution=384 x 2882022.01 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | |
| HRNetNetwork=HRNetW48, Input Size=384 x 2882023.03 | 75.5 | — | — | — | — | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetBackbone=HR48, Size=384×2882023.03 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | 80.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNet + UDPBackbone=HRNet-W32, Input size=256 x 192, Paradigm=Top-down2019.11 | 75.2 | 92.4 | 82.9 | 72 | 80.8 | 80.4 | — | — | — | — | — | — | — | — | — | — |