3D Human Pose Estimation on 3DPW
33.6PA-MPJPEMetricHMSR
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MetricHMSR2025.06 | 33.6 | 53 | — | — | 62.7 | — | — | — | — | — | — | — | |
| CameraHMR2025.06 | 35.1 | 56 | — | — | 65.9 | — | — | — | — | — | — | — | |
| PromptHMR2025.06 | 35.5 | 56.9 | — | — | 67.3 | — | — | — | — | — | — | — | |
| TRAM2025.06 | 35.6 | 59.3 | — | — | 69.6 | — | — | — | — | — | — | — | |
| WHAM2025.06 | 35.9 | 57.8 | — | — | 68.7 | — | — | — | — | — | — | — | |
| PersPose2025.06 | 39.1 | 60.1 | — | — | 72.4 | — | — | — | — | — | — | — | |
| ZeDOSupervision=Unsupervised, Number of hypotheses (S)=1, Number of joints (J)=17, 2D Input=GT2023.07 | 40.3 | — | — | — | — | 60.8 | — | — | — | — | — | — | |
| NIKIMapping Strategy=Twist-and-Swing2023.05 | 40.6 | 71.3 | — | — | 86.6 | — | — | — | — | — | — | — | |
| NIKI#Params=92.17, FPS=6.182024.06 | 41 | 71.7 | — | — | 86.9 | — | — | — | — | — | — | — | |
| NIKIMapping Strategy=One-Stage2023.05 | 41 | 71.7 | — | — | 86.9 | — | — | — | — | — | — | — | |
| HybrIK#Params=75.69, FPS=7.572024.06 | 41.8 | 71.6 | — | — | 82.3 | — | — | — | — | — | — | — | |
| HybrIKMethod Category=Pixel-aligned, Backbone=HRNet2023.05 | 41.8 | 72.9 | — | — | 88.6 | — | — | — | — | — | — | — | |
| CLIFF + PROMorthogonalization=PROM2023.12 | 42 | 67.6 | — | — | — | — | 79.2 | — | — | — | — | — | |
| 3DHR-CoBackbone=SPIN, Noise Level (sigma)=50, Intermediate Label Data Extractor=MPN2023.10 | 42.1 | 63.87 | — | — | — | — | — | -34.38 | -18.09 | — | — | — | |
| 3DHR-CoBackbone=RSN, Noise Level (sigma)=50, Intermediate Label Data Extractor=MPN2023.10 | 42.1 | 64.77 | — | — | — | — | — | -33.74 | -17.08 | — | — | — | |
| 3DHR-CoBackbone=SPIN, Noise Level (sigma)=35, Intermediate Label Data Extractor=MPN2023.10 | 42.11 | 63.72 | — | — | — | — | — | -34.53 | -18.08 | — | — | — | |
| 3DHR-CoBackbone=RSN, Noise Level (sigma)=35, Intermediate Label Data Extractor=MPN2023.10 | 42.11 | 64.91 | — | — | — | — | — | -33.6 | -17.07 | — | — | — | |
| 3DHR-CoBackbone=SPIN, Noise Level (sigma)=65, Intermediate Label Data Extractor=MPN2023.10 | 42.54 | 63.89 | — | — | — | — | — | -34.36 | -17.65 | — | — | — | |
| 3DHR-CoBackbone=RSN, Noise Level (sigma)=65, Intermediate Label Data Extractor=MPN2023.10 | 42.54 | 65.19 | — | — | — | — | — | -33.32 | -16.64 | — | — | — | |
| CLIFF2023.12 | 43 | 69 | — | — | — | — | 81.2 | — | — | — | — | — | |
| CLIFF#Params=78.89, FPS=22.422024.06 | 43 | 69 | — | — | 81.2 | — | — | — | — | — | — | — | |
| Human3R2025.06 | 44.1 | 72.1 | — | — | 84.9 | — | — | — | — | — | — | — | |
| Semantics-aware Test-time AdaptationBackbone=HMR2.02025.02 | 44.3 | 71.2 | — | — | — | — | — | — | — | — | — | — | |
| HMR2.0a2025.06 | 44.4 | 69.8 | — | — | 82.2 | — | — | — | — | — | — | — | |
| HiPARTInput type=2D keypoints2026.05 | 44.5 | 112.4 | — | — | — | — | — | — | — | — | — | — | |
| Again-Poseinput=video-based2026.06 | 44.5 | 70.7 | — | — | — | — | — | — | — | — | — | — | |
| FastMetro#Params=153.74, FPS=16.042024.06 | 44.6 | 73.5 | — | — | 84.1 | — | — | — | — | — | — | — | |
| PointHMR#Params=59.09, FPS=15.382024.06 | 44.9 | 73.9 | — | — | 85.5 | — | — | — | — | — | — | — | |
| SAM 3D BodyInput type=Monocular video2026.05 | 45 | 95 | — | — | — | — | — | — | — | — | — | — | |
| OSXModel=SMPLX, Predict parts=Body,face,hand, FPS=202026.01 | 45.1 | 74.7 | — | — | — | — | — | — | — | — | — | — | |
| HybrIKMethod Category=Pixel-aligned, Backbone=ResNet2023.05 | 45.1 | 76.2 | — | — | 89.1 | — | — | — | — | — | — | — | |
| Ours (EHM-s)Model=EHM-s, Predict parts=Body,face,hand, FPS=1002026.01 | 45.3 | 71.3 | — | — | — | — | — | — | — | — | — | — | |
| Mesh GraphormerUses 3D supervision=true2021.05 | 45.6 | 74.7 | — | — | — | — | — | — | — | — | — | — | |
| Graphormer#Params=226.21, FPS=14.422024.06 | 45.6 | 74.7 | — | — | 87.7 | — | — | — | — | — | — | — | |
| Lifelong Domain Adaptive 3D HPE frameworkTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 45.6 | 81.7 | — | — | — | — | — | — | — | — | — | — | |
| Mesh GraphormerMethod Category=Pixel-aligned2023.05 | 45.6 | 74.7 | — | — | 87.7 | — | — | — | — | — | — | — | |
| SMPLest-XModel=SMPLX, Predict parts=Body,face,hand, FPS=202026.01 | 45.8 | 74.8 | — | — | — | — | — | — | — | — | — | — | |
| MotionAGFormerInput type=2D keypoints2026.05 | 45.8 | 108.6 | — | — | — | — | — | — | — | — | — | — | |
| Multi-HMRModel=SMPLX, Predict parts=Body,face,hand, FPS=102026.01 | 45.9 | 78 | — | — | — | — | — | — | — | — | — | — | |
| MotionBERTInput type=2D keypoints2026.05 | 46.3 | 110.2 | — | — | — | — | — | — | — | — | — | — | |
| PARE#Params=32.86, FPS=20.482024.06 | 46.5 | 74.5 | — | — | 88.6 | — | — | — | — | — | — | — | |
| PAREMethod Category=Regression2023.05 | 46.5 | 74.5 | — | — | 88.6 | — | — | — | — | — | — | — | |
| CameraHMRInput type=Monocular video2026.05 | 46.5 | 98.3 | — | — | — | — | — | — | — | — | — | — | |
| MeshPoseBackbone=HRNet32, #Params=46.29, FPS=28.662024.06 | 46.73 | 76.08 | — | — | 92.7 | — | — | — | — | — | — | — | |
| CycleAdaptBackbone=HMR2.02025.02 | 46.8 | 73.7 | — | — | — | — | — | — | — | — | — | — | |
| Lifelong Domain Adaptive 3D HPE frameworkTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DPW → 3DHP2025.12 | 46.9 | 83.7 | — | — | — | — | — | — | — | — | — | — | |
| Semantics-aware Test-time AdaptationBackbone=HMR, Pre-training Dataset=Human3.6M2025.02 | 47.2 | 76.4 | — | — | — | — | 94 | — | — | — | — | — | |
| METROUses 3D supervision=true2021.05 | 47.9 | 77.1 | — | — | — | — | — | — | — | — | — | — | |
| Metro#Params=243.07, FPS=15.472024.06 | 47.9 | 77.1 | — | — | 88.2 | — | — | — | — | — | — | — | |
| METROMethod Category=Regression2023.05 | 47.9 | 77.1 | — | — | 88.2 | — | — | — | — | — | — | — | |
| PARE + PEFine-tuned with PoseExaminer=true2023.03 | 48.01 | — | — | — | 94.86 | — | — | — | — | — | — | — | |
| H-FlowInput type=Monocular video2026.05 | 48.7 | 91.5 | — | — | — | — | — | — | — | — | — | — | |
| HybrIK3D Supervision=true2021.10 | 48.8 | 80 | — | — | — | — | — | — | — | — | — | — | |
| HybrIK2023.12 | 48.8 | 80 | — | — | — | — | 94.5 | — | — | — | — | — | |
| PromptHMRInput type=Monocular video2026.05 | 48.9 | 102.7 | — | — | — | — | — | — | — | — | — | — | |
| MeshPoseSBackbone=ResNet50, #Params=45.37, FPS=124.282024.06 | 48.97 | 80.03 | — | — | 97.96 | — | — | — | — | — | — | — | |
| HMR-EFT + PEFine-tuned with PoseExaminer=true2023.03 | 49.15 | — | — | — | 101.43 | — | — | — | — | — | — | — | |
| PoseDA-LLTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 49.4 | 87.6 | — | — | — | — | — | — | — | — | — | — | |
| MeshPoseXSBackbone=MBNet140, #Params=21.25, FPS=124.212024.06 | 49.71 | 79.15 | — | — | 96.49 | — | — | — | — | — | — | — | |
| Adaptpose-LLTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 50.2 | 88.2 | — | — | — | — | — | — | — | — | — | — | |
| CycleAdapt-LLTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 50.2 | 88.3 | — | — | — | — | — | — | — | — | — | — | |
| RMT-PoseTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 50.4 | 88.5 | — | — | — | — | — | — | — | — | — | — | |
| CoTTA-PoseTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DHP → 3DPW2025.12 | 50.5 | 88.7 | — | — | — | — | — | — | — | — | — | — | |
| PoseMoE2025.12 | 50.6 | 76.8 | — | — | — | — | — | — | — | — | — | — | |
| GLoTinput=video-based, trained on 3DPW=true2026.06 | 50.6 | 80.7 | — | — | — | — | — | — | — | — | — | — | |
| PAREFine-tuned with PoseExaminer=false2023.03 | 50.78 | — | — | — | 102.27 | — | — | — | — | — | — | — | |
| KAMAMethod Category=Pixel-aligned2023.05 | 51.1 | — | — | — | 97 | — | — | — | — | — | — | — | |
| HMR-EFTtraining_data=3DPW2021.04 | 51.7 | 85.3 | — | — | — | — | — | — | — | — | — | — | |
| TUCHEXtraining_data=3DPW2021.04 | 51.7 | 82.3 | — | — | — | — | — | — | — | — | — | — | |
| HMR-EFTFine-tuned with PoseExaminer=false2023.03 | 51.88 | — | — | — | 107.88 | — | — | — | — | — | — | — | |
| VIBEUses 3D supervision=true2021.05 | 51.9 | 82 | — | — | — | — | — | — | — | — | — | — | |
| KTPFormerVenue=CVPR’242025.12 | 52.7 | 81 | — | — | — | — | — | — | — | — | — | — | |
| HMR2.0Backbone=HMR2.02025.02 | 53.3 | 80.2 | — | — | — | — | — | — | — | — | — | — | |
| ROMPMethod Category=Regression2023.05 | 53.3 | 85.5 | — | — | 103.1 | — | — | — | — | — | — | — | |
| HierarchicalProbabilistic3DHuman3D Supervision=false, 2D joint detector=HRNet-W482021.10 | 53.6 | 84.9 | — | 73 | — | — | — | — | — | — | — | — | |
| CycleAdaptBackbone=HMR, Pre-training Dataset=Human3.6M2025.02 | 53.8 | 87.7 | — | — | — | — | 105.7 | — | — | — | — | — | |
| PoseDA-LLTime (t)=2, Adaptation Sequence=H3.6M (Source) → 3DPW → 3DHP2025.12 | 53.8 | 91.5 | — | — | — | — | — | — | — | — | — | — | |
| MixSTEVenue=CVPR’222025.12 | 54.3 | 87.4 | — | — | — | — | — | — | — | — | — | — | |
| SPIN + PEFine-tuned with PoseExaminer=true2023.03 | 54.34 | — | — | — | 103.94 | — | — | — | — | — | — | — | |
| STCFormerVenue=CVPR’232025.12 | 54.5 | 84.9 | — | — | — | — | — | — | — | — | — | — | |
| DaNet3D Supervision=true2021.10 | 54.8 | 85.5 | — | 76.4 | — | — | — | — | — | — | — | — | |
| DaNet#Params=102.25, FPS=11.952024.06 | 54.8 | 85.5 | — | — | 110.8 | — | — | — | — | — | — | — | |
| HumanSplatHMRSMPL input source=HMR2.0 [4]2026.05 | 55.1 | 72.63 | — | — | — | — | — | — | — | — | — | 196.95 | |
| GLA-GCNVenue=ICCV’232025.12 | 55.8 | 91.6 | — | — | — | — | — | — | — | — | — | — | |
| PoseFormerV2Venue=CVPR’232025.12 | 56.2 | 91.1 | — | — | — | — | — | — | — | — | — | — | |
| HMR2.0SMPL input source=HMR2.0 [4]2026.05 | 56.26 | 75.64 | — | — | — | — | — | — | — | — | — | 239.27 | |
| GARTSMPL input source=HMR2.0 [4]2026.05 | 56.27 | 75.64 | — | — | — | — | — | — | — | — | — | 239.27 | |
| VIBE2023.12 | 56.5 | 93.5 | — | — | — | — | 113.4 | — | — | — | — | — | |
| VIBEinput=video-based2026.06 | 56.5 | 93.5 | — | — | — | — | — | — | — | — | — | — | |
| MHFormerVenue=CVPR’222025.12 | 56.7 | 92.3 | — | — | — | — | — | — | — | — | — | — | |
| IPMAN-R2023.03 | 57.1 | 96.8 | — | — | — | — | — | — | — | — | — | — | |
| SPINFine-tuned with PoseExaminer=false2023.03 | 57.54 | — | — | — | 111.12 | — | — | — | — | — | — | — | |
| I2L-MeshNet3D Supervision=true2021.10 | 57.7 | 93.2 | — | 77.5 | — | — | — | — | — | — | — | — | |
| I2LMeshNetUses 3D supervision=true2021.05 | 57.7 | 93.2 | — | — | — | — | — | — | — | — | — | — | |
| HMR 2.02024.05 | 58.4 | 91 | — | — | — | — | — | — | — | — | 9.2 | — | |
| I2LMethod Category=Pixel-aligned2023.05 | 58.6 | 93.2 | — | — | — | — | — | — | — | — | — | — | |
| ChatHuman2024.05 | 58.7 | 91.3 | — | — | — | — | — | — | — | — | 9.2 | — | |
| Pose2Mesh3D Supervision=true2021.10 | 58.9 | 89.2 | — | — | — | — | — | — | — | — | — | — | |
| Pose2MeshUses 3D supervision=true2021.05 | 58.9 | 89.2 | — | — | — | — | — | — | — | — | — | — | |
| PyMaf#Params=45.18, FPS=35.572024.06 | 58.9 | 92.8 | — | — | 110.1 | — | — | — | — | — | — | — |