Human Pose Estimation on COCO 2017 (test-dev)
78.3APPCT
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| PCTBackbone=Swin-Huge, Input size=256 x 256, GFLOPs=118.2, Speed (fps)=31.72023.03 | 78.3 | — | — | 92.9 | 85.9 | — | — | — | — | — | |
| ViTPoseBackbone=ViT-Huge, Input size=256 x 192, GFLOPs=122.9, Speed (fps)=21.82023.03 | 78.1 | — | — | 93.3 | 85.7 | — | — | — | — | — | |
| PCTBackbone=Swin-Large, Input size=256 x 256, GFLOPs=34.1, Speed (fps)=76.42023.03 | 77.4 | — | — | 92.9 | 85.2 | — | — | — | — | — | |
| ViTPoseBackbone=ViT-Large, Input size=256 x 192, GFLOPs=59.8, Speed (fps)=40.52023.03 | 77.3 | — | — | 93.1 | 85.3 | — | — | — | — | — | |
| HRNet-W48Approach=Top-down, Extra data=true2019.08 | 77 | 73.4 | 83.1 | 92.7 | 84.5 | 82 | — | — | — | — | |
| HRNetV1 + extra dataApproach=Top-down, Backbone=HRNetV1-W48, Input size=384 x 288, #Params=63.6M, GFLOPS=32.9, Extra data=true2019.08 | 77 | 73.4 | 83.1 | 92.7 | 84.5 | 82 | — | — | — | — | |
| EvoPose2D-L*PT (Pre-training)=N, Input Size=512x384, Params (M)=14.7, FLOPs (G)=17.7, Size (MB)=60.6, FPS (GPU)=15.9, Post-processing=PoseFix2020.11 | 76.8 | 73.5 | 82.5 | 92.5 | 84.3 | 81.7 | — | — | — | — | |
| HRNet-W48*PT (Pre-training)=Y, Input Size=384x288, Params (M)=63.6, FLOPs (G)=35.3, Size (MB)=259, FPS (GPU)=16.2, Post-processing=PoseFix2020.11 | 76.7 | 73.1 | 82.6 | 92.6 | 84.1 | 81.5 | — | — | — | — | |
| UDPBackbone=HRNet-W48, Input size=384 x 288, GFLOPs=35.5, Speed (fps)=67.92023.03 | 76.5 | — | — | 92.7 | 84 | — | — | — | — | — | |
| PCTBackbone=Swin-Base, Input size=256 x 256, GFLOPs=15.2, Speed (fps)=115.12023.03 | 76.5 | — | — | 92.5 | 84.7 | — | — | — | — | — | |
| DARKBackbone=HRNet-W48, Input size=384 x 288, GFLOPs=35.5, Speed (fps)=62.12023.03 | 76.2 | — | — | 92.5 | 83.6 | — | — | — | — | — | |
| HRFormerBackbone=HRFormer-B, Input size=384 x 288, GFLOPs=29.1, Speed (fps)=25.22023.03 | 76.2 | — | — | 92.7 | 83.8 | — | — | — | — | — | |
| DARKBackbone=HRNet-W48, Input size=384 × 288, #Params (M)=63.6, GFLOPs=32.92022.04 | 76.2 | 72.5 | 82.4 | 92.5 | 83.6 | 81.1 | — | — | — | — | |
| UDPBackbone=HRNet-W32, Input size=384 × 288, #Params (M)=28.7, GFLOPs=16.12022.04 | 76.1 | 72.8 | 82 | 92.5 | 83.5 | 81.3 | — | — | — | — | |
| SimCCBackbone=HRNet-W48, Input size=384 x 288, GFLOPs=32.9, Speed (fps)=71.42023.03 | 76 | — | — | 92.4 | 83.5 | — | — | — | — | — | |
| TokenPoseBackbone=HRNet-W48, Input size=256 x 192, GFLOPs=22.1, Speed (fps)=52.92023.03 | 75.9 | — | — | 92.3 | 83.4 | — | — | — | — | — | |
| EvoPose2D-LPT (Pre-training)=N, Input Size=512x384, Params (M)=14.7, FLOPs (G)=17.7, Size (MB)=60.6, FPS (GPU)=15.92020.11 | 75.7 | 72.2 | 81.5 | 91.9 | 83.1 | 81.7 | — | — | — | — | |
| HRNet-W48Approach=Top-down2019.08 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | 80.5 | — | — | — | — | |
| HRNetV1Approach=Top-down, Backbone=HRNetV1-W48, Input size=384 x 288, #Params=63.6M, GFLOPS=32.92019.08 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | 80.5 | — | — | — | — | |
| HRNet-W48PT (Pre-training)=Y, Input Size=384x288, Params (M)=63.6, FLOPs (G)=35.3, Size (MB)=259, FPS (GPU)=16.22020.11 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | 80.5 | — | — | — | — | |
| HRNet-W48Method Type=Top-down, Backbone=HRNet-W482021.10 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | — | — | — | — | — | |
| HRNetParadigm=Top-down2021.11 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | 80.5 | — | — | — | — | |
| HRNetBackbone=HRNet-W482021.09 | 75.5 | 71.9 | 81.5 | 92.5 | 83.3 | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48, Input size=384 x 288, GFLOPs=35.5, Speed (fps)=75.52023.03 | 75.5 | — | — | 92.7 | 83.3 | — | — | — | — | — | |
| SimBa.Backbone=Swin-Base, Input size=256 x 256, GFLOPs=16.6, Speed (fps)=74.42023.03 | 75.4 | — | — | 93 | 84.1 | — | — | — | — | — | |
| ViTPoseBackbone=ViT-Base, Input size=256 x 192, GFLOPs=17.9, Speed (fps)=113.52023.03 | 75.1 | — | — | 92.5 | 83.1 | — | — | — | — | — | |
| TransPoseBackbone=HRNet-W48, Input size=256 x 192, GFLOPs=21.8, Speed (fps)=56.72023.03 | 75 | — | — | 92.2 | 82.3 | — | — | — | — | — | |
| HRNetV1Approach=Top-down, Backbone=HRNetV1-W32, Input size=384 x 288, #Params=28.5M, GFLOPS=162019.08 | 74.9 | 71.3 | 80.9 | 92.5 | 82.8 | 80.1 | — | — | — | — | |
| HRNetBackbone=HRNet-W32, Input size=384 × 288, #Params (M)=28.5, GFLOPs=162022.04 | 74.9 | 71.3 | 80.9 | 92.5 | 82.8 | 80.1 | — | — | — | — | |
| S-ViPNAS-HRNetW32Image Size=256 x 192, #Params=16.3M, GFLOPS=5.642021.05 | 73.9 | 70.5 | 79.5 | 91.7 | 82 | 80.4 | — | — | — | — | |
| SimpleBaselineApproach=Top-down2019.08 | 73.7 | 70.3 | 80 | 91.9 | 81.1 | 79 | — | — | — | — | |
| SimpleBaselineApproach=Top-down, Backbone=ResNet-152, Input size=384 x 288, #Params=68.6M, GFLOPS=35.62019.08 | 73.7 | 70.3 | 80 | 91.9 | 81.1 | 79 | — | — | — | — | |
| SB (R-152)PT (Pre-training)=Y, Input Size=384x288, Params (M)=68.6, FLOPs (G)=28.1, Size (MB)=277, FPS (GPU)=24.92020.11 | 73.7 | 70.3 | 80 | 91.9 | 81.1 | 79 | — | — | — | — | |
| SimpleBaselineMethod Type=Top-down2021.10 | 73.7 | 70.3 | 80 | 91.9 | 81.1 | — | — | — | — | — | |
| SimpleBaselineParadigm=Top-down2021.11 | 73.7 | 70.3 | 80.8 | 91.9 | 81.1 | 79 | — | — | — | — | |
| SimBa.Backbone=ResNet-152, Input size=384 x 288, GFLOPs=28.7, Speed (fps)=76.32023.03 | 73.7 | — | — | 91.9 | 81.1 | — | — | — | — | — | |
| HRNet-W32Image Size=256 x 192, #Params=28.5M, GFLOPS=7.12021.05 | 73.5 | 70.1 | 79.1 | 91.6 | 81.7 | 80.1 | — | — | — | — | |
| CPN (ensemble)Approach=Top-down, Ensemble=true2019.08 | 73 | 69.5 | 78.1 | 91.7 | 80.9 | 79 | — | — | — | — | |
| CPN (ensemble)Approach=Top-down, Backbone=ResNet-Inception, Input size=384 x 288, Ensemble=true2019.08 | 73 | 69.5 | 78.1 | 91.7 | 80.9 | 79 | — | — | — | — | |
| CFNApproach=Top-down2019.08 | 72.6 | 78.3 | 64.1 | 86.1 | 69.7 | — | — | — | — | — | |
| CFNApproach=Top-down2019.08 | 72.6 | 78.3 | 64.1 | 86.1 | 69.7 | — | — | — | — | — | |
| CFNMethod Type=Top-down2021.10 | 72.6 | 78.3 | 64.1 | 86.1 | 69.7 | — | — | — | — | — | |
| RMPEApproach=Top-down2019.08 | 72.3 | 68 | 78.6 | 89.2 | 79.1 | — | — | — | — | — | |
| RMPEApproach=Top-down, Backbone=PyraNet [165], Input size=320 x 256, #Params=28.1M, GFLOPS=26.72019.08 | 72.3 | 68 | 78.6 | 89.2 | 79.1 | — | — | — | — | — | |
| PoseNFS-3Image Size=384 x 288, #Params=15.8M, GFLOPS=14.82021.05 | 72.3 | 68.4 | 79.2 | 90.9 | 79.5 | 77.9 | — | — | — | — | |
| RMPEMethod Type=Top-down2021.10 | 72.3 | 68 | 78.6 | 86.1 | 79.1 | — | — | — | — | — | |
| CPNApproach=Top-down2019.08 | 72.1 | 68.7 | 77.2 | 91.4 | 80 | 78.5 | — | — | — | — | |
| CPNApproach=Top-down, Backbone=ResNet-Inception, Input size=384 x 2882019.08 | 72.1 | 68.7 | 77.2 | 91.4 | 80 | 78.5 | — | — | — | — | |
| CPNPT (Pre-training)=Y, Input Size=384x2882020.11 | 72.1 | 68.7 | 77.2 | 91.4 | 80 | 78.5 | — | — | — | — | |
| CPNMethod Type=Top-down2021.10 | 72.1 | 68.7 | 77.2 | 91.4 | 80 | — | — | — | — | — | |
| CPNBackbone=ResNet-Inception, Input size=384 × 2882022.04 | 72.1 | 68.7 | 77.2 | 91.4 | 80 | 78.5 | — | — | — | — | |
| PRTRBackbone=HRNet-W32, Input size=384 x 288, GFLOPs=21.6, Speed (fps)=872023.03 | 71.7 | — | — | 90.6 | 79.6 | — | — | — | — | — | |
| ScaleNASBackbone=ScaleNet-P4(Ours), Input size=640, #Params=64.4M, GFLOPS=141.5, Multi-scale test=true2020.11 | 71.6 | — | — | — | — | — | — | — | — | — | |
| CenterGroupBackbone=HRNet-W48, Grouping=Attention, Bipartite matching solvers=false, Input size=640, # Params=65.5M, Multi-scale test=true2021.10 | 71.4 | 67.2 | 77.5 | 90.4 | 78.1 | — | — | — | — | — | |
| CenterGroupMethod Type=Bottom-up, Backbone=HrHRNet-W48, multi-scale testing=true2021.10 | 71.4 | 67.2 | 77.5 | 90.4 | 78.1 | — | — | — | — | — | |
| DEKRtesting_scale=multi-scale, Input size=640, Backbone=HRNet-W482021.04 | 71 | 67.1 | 76.9 | 89.2 | 78 | 76.7 | 71.5 | 83.9 | — | — | |
| DEKRParadigm=Bottom-up2021.11 | 71 | 67.1 | 76.9 | 89.2 | 78 | 76.7 | 71.5 | 83.9 | 93.2 | 83 | |
| LOGO-CAPBackbone=HRNet-W482021.09 | 70.8 | 66.7 | 77 | 89.7 | 77.8 | — | — | — | — | — | |
| Dite-HRNet-30Backbone=Dite-HRNet-30, Input size=384 × 288, #Params (M)=1.8, GFLOPs=0.72022.04 | 70.6 | 67.4 | 76.1 | 90.8 | 78.2 | 76.4 | — | — | — | — | |
| HigherHRNet-W48Approach=Bottom-up, Multi-scale test=true2019.08 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | 74.9 | — | — | — | — | |
| HigherHRNetBackbone=HRNet-W48, Input size=640, #Params=63.8M, GFLOPs=154.3, multi-scale test=true, refinement=false2019.08 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | — | — | — | — | — | |
| HigherHRNetBackbone=HRNet-W48, Input size=640, #Params=63.8M, GFLOPS=154.3, Multi-scale test=true2020.11 | 70.5 | — | — | — | — | — | — | — | — | — | |
| HrHRNet-W48 + AEtesting_scale=multi-scale, Input size=6402021.04 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | — | — | — | — | — | |
| HrHRNet-W48Backbone=HRNet-W48, Grouping=Greedy decoding w/ optimization, Bipartite matching solvers=true, Input size=640, # Params=63.8M, Multi-scale test=true2021.10 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | — | — | — | — | — | |
| HrHRNet-W48Method Type=Bottom-up, Backbone=HrHRNet-W48, multi-scale testing=true2021.10 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | — | — | — | — | — | |
| HigherHRNetParadigm=Bottom-up2021.11 | 70.5 | 66.6 | 75.8 | 89.3 | 77.2 | — | — | — | — | — | |
| S-ViPNAS-Res50Image Size=256 x 192, #Params=13.5M, GFLOPS=1.442021.05 | 70.3 | 67.3 | 75.5 | 90.7 | 78.8 | 77.3 | — | — | — | — | |
| CenterGroupBackbone=HRNet-W32, Grouping=Attention, Bipartite matching solvers=false, Input size=512, # Params=30.3M, Multi-scale test=true2021.10 | 70.3 | 65.4 | 77.5 | 90 | 76.9 | — | — | — | — | — | |
| CenterGroupMethod Type=Bottom-up, Backbone=HrHRNet-W32, multi-scale testing=true2021.10 | 70.3 | 65.4 | 77.5 | 90 | 76.9 | — | — | — | — | — | |
| DEKRtesting_scale=single-scale, Input size=640, Backbone=HRNet-W482021.04 | 70 | 65.7 | 76.9 | 89.4 | 77.3 | 75.4 | 69.7 | 83.2 | — | — | |
| SBL-50Image Size=256 x 192, #Params=34.0M, GFLOPS=8.92021.05 | 70 | 66.8 | 75.8 | 90.9 | 77.9 | 75.6 | — | — | — | — | |
| DEKRBackbone=HRNet-W48, Rescoring=true2021.09 | 70 | 65.7 | 76.9 | 89.4 | 77.3 | — | — | — | — | — | |
| SimpleBaselineBackbone=ResNet-50, Input size=256 × 192, #Params (M)=34, GFLOPs=8.92022.04 | 70 | 66.8 | 75.8 | 90.9 | 77.9 | 75.6 | — | — | — | — | |
| DEKRtesting_scale=multi-scale, Input size=512, Backbone=HRNet-W322021.04 | 69.8 | 65.2 | 76.5 | 89 | 76.6 | 75.1 | 69.5 | 82.8 | — | — | |
| Lite-HRNetBackbone=Lite-HRNet-302021.09 | 69.7 | 66.9 | 75 | 90.7 | 77.5 | — | — | — | — | — | |
| Lite-HRNet-30Backbone=Lite-HRNet-30, Input size=384 × 288, #Params (M)=1.8, GFLOPs=0.72022.04 | 69.7 | 66.9 | 75 | 90.7 | 77.5 | 75.4 | — | — | — | — | |
| MultiPoseNetApproach=Bottom-up2019.08 | 69.6 | 65 | 76.3 | 86.3 | 76.6 | 73.5 | — | — | — | — | |
| CenterGroupBackbone=HRNet-W48, Grouping=Attention, Bipartite matching solvers=false, Input size=640, # Params=65.5M, Multi-scale test=false2021.10 | 69.6 | 64.9 | 76.3 | 89.7 | 76 | — | — | — | — | — | |
| CenterGroupMethod Type=Bottom-up, Backbone=HrHRNet-W48, multi-scale testing=false2021.10 | 69.6 | 64.9 | 76.3 | 89.7 | 76 | — | — | — | — | — | |
| DEKRBackbone=HRNet-W48, Rescoring=false2021.09 | 69.3 | 65.3 | 76.4 | 89.1 | 76.7 | — | — | — | — | — | |
| PersonLabApproach=Bottom-up, Multi-scale test=true2019.08 | 68.7 | 64.1 | 75.5 | 89 | 75.4 | 75.4 | — | — | — | — | |
| PersonLabBackbone=ResNet-152, Input size=1401, #Params=68.7M, GFLOPs=405.5, multi-scale test=true, refinement=false2019.08 | 68.7 | 64.1 | 75.5 | 89 | 75.4 | — | — | — | — | — | |
| PersonLabApproach=Bottom-up2019.08 | 68.7 | 64.1 | 75.5 | 89 | 75.4 | 75.4 | — | — | — | — | |
| PersonLabBackbone=ResNet-152, Input size=1401, #Params=68.7M, GFLOPS=405.5, Multi-scale test=true2020.11 | 68.7 | — | — | — | — | — | — | — | — | — | |
| PersonLabtesting_scale=multi-scale, Input size=14012021.04 | 68.7 | 64.1 | 75.5 | 89 | 75.4 | 75.4 | 69.7 | 83 | — | — | |
| Point-Set Anchorstesting_scale=multi-scale, Input size=6402021.04 | 68.7 | 64.8 | 75.3 | 89.9 | 76.3 | 74.8 | 69.6 | 82.1 | — | — | |
| PersonLabBackbone=ResNet152, Grouping=Greedy decoding, Bipartite matching solvers=false, Input size=1401, # Params=68.7M, Multi-scale test=true2021.10 | 68.7 | 66.6 | 75.8 | 89 | 75.4 | — | — | — | — | — | |
| PersonLabMethod Type=Bottom-up, multi-scale testing=true2021.10 | 68.7 | 66.6 | 75.8 | 89 | 75.4 | — | — | — | — | — | |
| PersonLabParadigm=Bottom-up2021.11 | 68.7 | 64.1 | 75.5 | 89 | 75.4 | 75.4 | 69.7 | 83 | 92.7 | 81.2 | |
| G-RMIApproach=Top-down, Extra data=true2019.08 | 68.5 | 65.8 | 73.3 | 87.1 | 75.5 | 73.3 | — | — | — | — | |
| G-RMI + extra dataApproach=Top-down, Backbone=ResNet-101, Input size=353 x 257, #Params=42.6M, GFLOPS=57, Extra data=true2019.08 | 68.5 | 65.8 | 73.3 | 87.1 | 75.5 | 73.3 | — | — | — | — | |
| SimplePoseBackbone=IMHN, Multi-scale inference=true2021.09 | 68.5 | 66.4 | 71.9 | 86.7 | 74.9 | — | — | — | — | — | |
| HigherHRNetBackbone=HRNet-W48, Input size=640, #Params=63.8M, GFLOPs=154.3, multi-scale test=false, refinement=false2019.08 | 68.4 | 64.4 | 74.2 | 88.2 | 75.1 | — | — | — | — | — | |
| HrHRNet-W48 + AEtesting_scale=single-scale, Input size=6402021.04 | 68.4 | 64.4 | 74.2 | 88.2 | 75.1 | — | — | — | — | — | |
| HigherHRNetBackbone=HRNet-W48, Grouping=Greedy decoding w/ optimization, Bipartite matching solvers=true, Input size=640, # Params=63.8M, Multi-scale test=false2021.10 | 68.4 | 64.4 | 74.2 | 88.2 | 75.1 | — | — | — | — | — | |
| HrHRNet-W48Method Type=Bottom-up, Backbone=HrHRNet-W482021.10 | 68.4 | 64.4 | 74.2 | 88.2 | 75.1 | — | — | — | — | — | |
| AEBackbone=HrHRNet-W482021.09 | 68.4 | 64.4 | 74.2 | 88.2 | 75.1 | — | — | — | — | — | |
| Dite-HRNet-18Backbone=Dite-HRNet-18, Input size=384 × 288, #Params (M)=1.1, GFLOPs=0.42022.04 | 68.4 | 65.2 | 73.8 | 89.9 | 75.8 | 74.4 | — | — | — | — | |
| LOGO-CAPBackbone=HRNet-W322021.09 | 68.2 | 62.8 | 76 | 88.7 | 74.9 | — | — | — | — | — | |
| SimplePosetesting_scale=multi-scale, Input size=5122021.04 | 68.1 | 66.8 | 70.5 | — | — | 72.1 | — | — | — | — |