Human Mesh Recovery on 3DPW
35.6PA-MPJPETRAM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TRAMProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 35.6 | 59.3 | — | — | 69.6 | — | |
| Patient4D2026.03 | 35.8 | 54.8 | — | — | 65.2 | — | |
| WHAMHuman Model Type=SMPL2025.11 | 35.9 | 57.8 | — | — | 68.7 | — | |
| WHAMProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 35.9 | 57.8 | — | — | 68.7 | — | |
| PromptHMR2026.03 | 36.6 | 58.7 | — | — | 69.4 | — | |
| FactorizedHMRProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 36.9 | 56.2 | — | — | 68.3 | — | |
| GVHMR2026.03 | 37 | 56.6 | — | — | 68.7 | — | |
| GVHMRProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 37.1 | 56.9 | — | — | 68.8 | — | |
| CoMotion2026.03 | 37.3 | 60 | — | — | 71.5 | — | |
| ReFitHuman Model Type=SMPL2025.11 | 38.2 | 57.6 | — | — | 67.6 | — | |
| HulkBackbone=ViT-L, Fine-tuning=false2023.12 | 38.5 | 66.3 | 77.4 | — | — | — | |
| CameraHMRHuman Model Type=SMPL2025.11 | 38.7 | 62.7 | — | — | 73.4 | — | |
| SKEL-CFHuman Model Type=SKEL2025.11 | 38.7 | 61.5 | — | — | 73.5 | — | |
| InterMeshBackbone=ViT-Base2026.05 | 39.8 | 60.7 | — | — | 71 | — | |
| HulkBackbone=ViT-B, Fine-tuning=false2023.12 | 39.9 | 67 | 79.8 | — | — | — | |
| ReFitProcessing type=Per-frame, Trained with 3DPW training set=true2026.05 | 40.5 | 65.3 | — | — | 75.1 | — | |
| HulkBackbone=ViT-L, Fine-tuning=true2023.12 | 40.6 | 68.3 | 79.9 | — | — | — | |
| MotionBERT (finetune) + HybrIKInput=video, T=162022.10 | 40.6 | 68.8 | — | — | 79.4 | — | |
| ProHMRNumber of hypotheses (n)=min2021.08 | 40.8 | — | — | — | — | — | |
| HulkBackbone=ViT-B, Fine-tuning=true2023.12 | 41.3 | 68.9 | 80.7 | — | — | — | |
| SAT-HMR2026.05 | 41.6 | 63.6 | — | — | 73.7 | — | |
| Multi-HMR2026.05 | 41.7 | 61.4 | — | — | 75.9 | — | |
| HybrIKProcessing type=Per-frame, Trained with 3DPW training set=true2026.05 | 41.8 | 71.6 | — | — | 82.3 | — | |
| HybrIKInput=image, T=12022.10 | 41.9 | 71.3 | — | — | 82.4 | — | |
| MotionBERT (finetune) + MAEDInput=video, T=162022.10 | 42.3 | 72.3 | — | — | 84.4 | — | |
| GENMOProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 42.9 | 62.3 | — | — | 82.8 | — | |
| CLIFFBackbone=ResNet-502023.12 | 43 | 69 | 81.2 | — | — | — | |
| CLIFFInput=image, T=12022.10 | 43 | 69 | — | — | 81.2 | — | |
| CLIFFHuman Model Type=SMPL2025.11 | 43 | 69 | — | — | 81.2 | — | |
| CLIFFProcessing type=Per-frame, Trained with 3DPW training set=true2026.05 | 43 | 69 | — | — | 81.2 | — | |
| TokenHMRHuman Model Type=SMPL2025.11 | 43.8 | 70.5 | — | — | 86 | — | |
| HMR2.0aHuman Model Type=SMPL2025.11 | 44.4 | 69.8 | — | — | 82.2 | — | |
| HMR 2.02026.03 | 44.4 | 69.8 | — | — | 82.2 | — | |
| HMR2.0Processing type=Per-frame, Trained with 3DPW training set=false2026.05 | 44.4 | 69.8 | — | — | 82.2 | — | |
| FastMETROBackbone=HRNet-W642023.12 | 44.6 | 73.5 | 84.1 | — | — | — | |
| FastMETROModel Scale=L, Backbone=HRNet-W642022.07 | 44.6 | 73.5 | 84.1 | — | — | — | |
| FastMETROInput=image, T=12022.10 | 44.6 | 73.5 | — | — | 84.1 | — | |
| POTTERParams (M)=16.3, MACS (G)=7.82023.03 | 44.8 | 75 | 87.4 | — | — | — | |
| VisDBBackbone=ResNet-502023.12 | 44.9 | 73.5 | 85.5 | — | — | — | |
| VisDBInput=image, T=12022.10 | 44.9 | 73.5 | — | — | 85.5 | — | |
| SMPLest-X2026.03 | 45.2 | 74.8 | — | — | 88.4 | — | |
| MeshGraphormerBackbone=HRNet-W642022.07 | 45.6 | 74.7 | 87.7 | — | — | — | |
| Mesh GraphormerInput=image, T=12022.10 | 45.6 | 74.7 | — | — | 87.7 | — | |
| MAEDInput=video, T=162022.10 | 45.7 | 79 | — | — | 93.3 | — | |
| PSVT2026.05 | 45.7 | 75.5 | — | — | 84.9 | — | |
| PAREBackbone=HRNet-W322023.12 | 46.5 | 74.5 | 88.6 | — | — | — | |
| PAREBackbone=HRNet-W322022.07 | 46.5 | 74.5 | 88.6 | — | — | — | |
| PAREInput=image, T=12022.10 | 46.5 | 74.5 | — | — | 88.6 | — | |
| CLIFFHuman Model Type=SMPL2025.11 | 46.6 | 72 | — | — | 85 | — | |
| DMAPS2026.02 | 46.68 | 69.31 | 82.61 | — | — | 7.14 | |
| MEAD2026.02 | 46.88 | 82.37 | 95.12 | — | — | 18.64 | |
| BEV2026.05 | 46.9 | 78.5 | — | — | 92.3 | — | |
| PMCE2026.02 | 47.05 | 74.76 | 86.31 | — | — | 6.96 | |
| MotionBERTBackbone=DSTformer2023.12 | 47.2 | 76.9 | 88.1 | — | — | — | |
| MotionBERT (finetune)Input=2D motion, T=162022.10 | 47.2 | 76.9 | — | — | 88.1 | — | |
| ROMP2026.05 | 47.3 | 76.6 | — | — | 93.4 | — | |
| METROBackbone=HRNet-W642023.12 | 47.9 | 77.1 | 88.2 | — | — | — | |
| METROBackbone=HRNet-W642022.07 | 47.9 | 77.1 | 88.2 | — | — | — | |
| METROParams (M)=229.2, MACS (G)=56.62023.03 | 47.9 | 77.1 | 88.2 | — | — | — | |
| METROInput=image, T=12022.10 | 47.9 | 77.1 | — | — | 88.2 | — | |
| ARTS2026.02 | 47.96 | 72.15 | 85.47 | — | — | 6.91 | |
| MotionBERT (finetune) + SPINInput=video, T=162022.10 | 48.2 | 79.6 | — | — | 92.8 | — | |
| FastMETROBackbone=ResNet-502023.12 | 48.3 | 77.9 | 90.6 | — | — | — | |
| FastMETROModel Scale=L, Backbone=ResNet-502022.07 | 48.3 | 77.9 | 90.6 | — | — | — | |
| FastMETROModel Scale=M, Backbone=ResNet-502022.07 | 48.4 | 78.5 | 91.2 | — | — | — | |
| Adaptive HybrIKBackbone=ResNet-342022.07 | 48.8 | 80 | — | — | 94.5 | — | |
| TCFormerResolution=224x2242022.04 | 49.3 | 80.6 | — | — | — | — | |
| SSFBackbone=ResNet-34, Synthetic data=w/o Synth2022.07 | 49.3 | 79.1 | — | — | 92.3 | — | |
| FastMETROModel Scale=S, Backbone=ResNet-502022.07 | 49.3 | 79.6 | 91.9 | — | — | — | |
| MotionBERT (scratch)Input=2D motion, T=162022.10 | 50.2 | 85.5 | — | — | 99.1 | — | |
| TRACEProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 50.9 | 79.1 | — | — | 95.4 | — | |
| PAREBackbone=ResNet-502022.07 | 51.2 | 84.3 | — | — | 101.2 | — | |
| THUNDRMACS (G)=252023.03 | 51.5 | 74.8 | 88 | — | — | — | |
| 3DCrowdNetInput=image, T=12022.10 | 51.5 | 81.7 | — | — | 98.3 | — | |
| 3DCrowdNet2026.05 | 51.5 | 81.7 | — | — | 98.3 | — | |
| GLoT2026.02 | 51.63 | 81.42 | 97.51 | — | — | 7.32 | |
| DSR2022.04 | 51.7 | 85.7 | — | — | — | — | |
| DSRBackbone=ResNet-502022.07 | 51.7 | 85.7 | 99.5 | — | — | — | |
| VIBEInput=video, T=162022.10 | 51.9 | 82.9 | — | — | 99.1 | — | |
| VIBEProcessing type=Temporal, Trained with 3DPW training set=true2026.05 | 51.9 | 82.9 | — | — | 98.4 | — | |
| MPS-NetInput=video, T=162022.10 | 52.1 | 84.3 | — | — | 99.7 | — | |
| EFT2022.04 | 52.2 | — | — | — | — | — | |
| PAREBackbone=ResNet-502022.07 | 52.3 | 82.9 | 99.7 | — | — | — | |
| ProHMRNumber of hypotheses (n)=252021.08 | 52.4 | — | — | — | — | — | |
| TCMRInput=video, T=162022.10 | 52.7 | 86.5 | — | — | 102.9 | — | |
| SmoothNetInput=video, T=322022.10 | 52.7 | 86.7 | — | — | — | — | |
| SPECHuman Model Type=SMPL2025.11 | 53.2 | 96.5 | — | — | 118.5 | — | |
| ROMPBackbone=HRNet-W322022.07 | 53.3 | 85.5 | 103.1 | — | — | — | |
| ROMPBackbone=ResNet-502022.07 | 53.5 | 89.3 | 105.6 | — | — | — | |
| HMR2.0bTraining Settings=8 × A100s, 1M Steps2024.07 | 54.3 | 81.3 | — | — | — | — | |
| HMR2.0bHuman Model Type=SMPL2025.11 | 54.3 | 81.3 | — | — | 93.1 | — | |
| ProHMRNumber of hypotheses (n)=102021.08 | 54.6 | — | — | — | — | — | |
| HambaTraining Settings=1 x A100, 300K Steps2024.07 | 54.7 | 81.7 | — | — | — | — | |
| HSMRHuman Model Type=SKEL2025.11 | 54.8 | 81.5 | — | — | — | — | |
| ROMPBackbone=ResNet-502022.07 | 54.9 | 91.3 | — | — | 108.3 | — | |
| ROMPInput=image, T=12022.10 | 54.9 | 91.3 | — | — | 108.3 | — | |
| Biggs et al. (NF)Number of hypotheses (n)=252021.08 | 55.6 | — | — | — | — | — | |
| Biggs et al. (NF)Number of hypotheses (n)=min2021.08 | 55.6 | — | — | — | — | — | |
| SLAHMRProcessing type=Temporal, Trained with 3DPW training set=false2026.05 | 55.9 | — | — | — | — | — | |
| VIBE2021.03 | 56.5 | 93.5 | 113.4 | — | — | — |