List your company

3D Human Pose Estimation on 3DPW (test)

29.9PA-MPJPECLIFF* + PROM
29.10434.47739.8545.223Mar 23, 2020Apr 8, 2021Apr 24, 2022May 11, 2023May 26, 2024Jun 11, 2025Jun 28, 2026

Evaluation results

MethodLinks
CLIFF* + PROMoptimization-based=true, 2D ground truth provided=true, refine_3D_rotation_by_2D_GT=true, orthogonalization_method=PROM2023.1229.949.5——56.9—————————————
Flexible Geometric GuidanceData=3D Only, Cross Domain=true, M=502026.0230.448.5————————————————
Jiang et al. (ZeDO)Data=3D Only, Cross Domain=true, M=502026.0230.654.8————————————————
CLIFF*optimization-based=true, 2D ground truth provided=true, refine_3D_rotation_by_2D_GT=true2023.1232.852.8——61.5—————————————
ExtPoseSingle=false2026.013454.2————————————————
GENMOCategory=Multimodal2026.0634.653.965.84.8——————————————
SparseNet2024.0735.456.6————————————————
TRAM2024.0735.659.3————————————————
TRAMCategory=V2M-only2026.0635.659.369.64.9——————————————
WHAMSingle=false2026.0135.957.8————————————————
WHAMCategory=V2M-only2026.0635.957.868.76.6——————————————
GVHMRCategory=V2M-only2026.0636.255.667.25——————————————
PromptHMR2024.0736.658.7————————————————
GVHMR2024.073756.6————————————————
WHAM2024.0737.559.8————————————————
ZeDOCross-domain evaluation=true, Optimization-based=true, Number of hypotheses (S)=1, Number of joints (J)=17, Ground truth 2D poses=true2023.0740.369.7————————————————
Jiang et al. (ZeDO)Data=3D Only, Cross Domain=true, M=12026.0240.369.7————————————————
DynaBOAProtocol=#PS2021.1140.465.5—————82——————————
NIKIMode=Temporal2023.0540.571.286.312.3——————————————
NIKIMethod category=image-based, Training Set=3DPW training set2023.1140.671.386.6———————————————
TARMethod category=video-based, Training Set=3DPW training set2023.1140.662.774.4———————————————
MotionBERTSingle=false2026.0140.668.8————————————————
NIKIMode=Frame-based2023.0540.671.386.615.1——————————————
ReFitMethod category=image-based, Training Set=3DPW training set2023.114165.8————————————————
MEGA2024.114167.5————————————————
SAMTCategory=Multimodal, Variant=w/o θ^l_r2026.0641.1255.8287.48.6——————————————
SAMTCategory=Multimodal2026.0641.169.9828.3——————————————
3DHR-CoBackbone=SPIN, Intr.-Pert.-Label=sigma_2 = 50, Intermediate Label Extractor=TCMR2023.1041.1463.92———————————-34.34-19.05———
3DHR-CoBackbone=SPIN, Intr.-Pert.-Label=sigma_1 = 35, Intermediate Label Extractor=TCMR2023.1041.2864.21———————————-34.04-18.91———
Ma et al.Cross-domain evaluation=false, Optimization-based=false, Ground truth 2D poses=true2023.0741.367.5————————————————
Virtual MarkerMethod category=image-based, Training Set=3DPW training set2023.1141.367.577.9———————————————
Ma et al.Data=Paired 2D-3D, Cross Domain=false2026.0241.367.5————————————————
3DHR-CoBackbone=SPIN, Intr.-Pert.-Label=sigma_3 = 65, Intermediate Label Extractor=TCMR2023.1041.3464.13———————————-34.12-18.85———
Multi-HMRtrained on 3DPW=true2026.0241.761.4————————————————
MultiHMRSingle=true2026.0141.761.4————————————————
HybrIKMethod category=image-based, Training Set=3DPW training set2023.1141.872.988.6———————————————
3DHR-CoBackbone=RSN, Intr.-Pert.-Label=sigma_3 = 65, Intermediate Label Extractor=TCMR2023.1041.965.14———————————-33.37-17.28———
3DHR-CoBackbone=RSN, Intr.-Pert.-Label=sigma_2 = 50, Intermediate Label Extractor=TCMR2023.1041.9764.82———————————-33.69-17.21———
INTMethod category=video-based, Training Set=3DPW training set2023.114275.687.9———————————————
3DHR-CoBackbone=RSN, Intr.-Pert.-Label=sigma_1 = 35, Intermediate Label Extractor=TCMR2023.1042.0564.83———————————-33.68-17.13———
FMPose3Dtrained on 3DPW=false, zero-shot evaluation=true, N=402026.0242.470.9————————————————
DynaBOAProtocol=#PV2021.1142.669—————84.4——————————
D&DTemporal Modeling=true2023.0542.773.788.67——————————————
CLIFFSingle=true2026.014369————————————————
CLIFF2024.074369————————————————
ZeDOCross-domain evaluation=true, Optimization-based=true, Number of hypotheses (S)=1, Number of joints (J)=14, Ground truth 2D poses=true2023.0743.176.6————————————————
SMPLer2024.1143.473.7————————————————
TokenHMR2024.1144.371————————————————
TokenHMR2024.0744.371————————————————
DynaBOAProtocol=#PH2021.1144.477.4—————89——————————
HMR2.02024.0744.469.8————————————————
HMR 2.0aMethod category=Frame-based, Non-parametric evaluation=false2023.0544.570————————————————
CLIFFMethod category=image-based, Training Set=3DPW training set2023.1144.573.5————————————————
HMR2.02024.1144.570————————————————
HMR2.0atrained on 3DPW=true2026.0244.570————————————————
HMR2.0Single=true2026.0144.570————————————————
Again-PoseTemporal Context=Single-frame, Trained on 3DPW=false2026.0644.570.7————————————————
FastMETROMethod category=image-based, Training Set=3DPW training set2023.1144.673.584.1———————————————
VisDBMethod category=image-based, Training Set=3DPW training set2023.1144.973.585.5———————————————
HybrIKMethod type=deterministic2023.054574.1————————————————
Li et al.Cross-domain evaluation=false, Optimization-based=false, Ground truth 2D poses=true2023.074574.1————————————————
HybrIKSingle=true2026.014574.1————————————————
PyMAFCategory=Single Stage, Backbone=HR48, training_on_3DPW=true2022.0745.374.287———————————————
PyMAFMethod category=image-based, Training Set=3DPW training set2023.1145.374.287———————————————
Mesh GraphormerProtocol=#PS, trained_on_3DPW_train_split=true2021.1145.674.7—————87.7——————————
GraphormerCategory=Single Stage, Backbone=HR64, training_on_3DPW=true2022.0745.674.787.7———————————————
Lin et al.CD (Cross-domain evaluation)=true, V (Video-based method)=false2023.0345.674.7————————————————
MAEDMethod category=video-based, Training Set=3DPW training set2023.1145.779.192.6———————————————
MAEDTemporal Modeling=true2023.0545.779.192.617.6——————————————
SPINProtocol=#PS, fine-tuned_on_3DPW_test_under_2D_supervision=true2021.114671.4—————97——————————
IVTFrames=5, transformer-based=true2022.0846—————————————————
PARE(HRNet-32)Protocol=#PS, trained_on_3DPW_train_split=true2021.1146.479.1—————94.2——————————
Kocabas et al. [25]Cross-domain evaluation=false, Optimization-based=false, Ground truth 2D poses=true2023.0746.474.7————————————————
PARECategory=Single Stage, Backbone=HR32, training_on_3DPW=true2022.0746.574.588.6———————————————
PAREMethod type=deterministic2023.0546.574.5————————————————
AdaptPoseCD (Cross-domain evaluation)=true, V (Video-based method)=false2023.0346.581.2————————————————
Gholami et al.Cross-domain evaluation=true, Optimization-based=false, Ground truth 2D poses=true2023.0746.581.2————————————————
PAREMethod category=image-based, Training Set=3DPW training set2023.1146.574.588.6———————————————
AdaptPoseType=Temporal2023.1146.5—————————————————
AdaptPosetrained on 3DPW=false, zero-shot evaluation=true2026.0246.581.2————————————————
Gholami et al.Data=Paired 2D-3D, Cross Domain=true2026.0246.581.2————————————————
BEDLAM-CLIFF2024.0746.672————————————————
PyMAFCategory=Single Stage, Backbone=ResNet-50, training_on_3DPW=true2022.0746.876.888.7———————————————
Hierarchical matrix-Fisher (Hierarchical)Number of Samples=25, Backbone=HRNet2021.104775.1————————————————
PyMAFCategory=Single Stage, Backbone=HR482022.0747.17891.3———————————————
PyMAF-XMethod category=Frame-based, Non-parametric evaluation=false2023.0547.178————————————————
ROMP(HRNet-32)Protocol=#PV, trained_on_3DPW_train_split=true2021.1147.376.7—————93.4——————————
METROProtocol=#PS, trained_on_3DPW_train_split=true2021.1147.977.1—————88.2——————————
Zolly2024.1147.976.2————————————————
BaselineCategory=Single Stage, Backbone=HR482022.074879.593.3———————————————
HybrIKProtocol=#PS2021.1148.880—————94.5——————————
HybrIKBody Model=SMPL2022.0648.880————————————————
HybrIKCategory=Multiple Stage2022.0748.88094.5———————————————
HyrbIKTemporal Context=Single-frame, Trained on 3DPW=false2026.0648.880————————————————
PyMAFCategory=Single Stage, Backbone=ResNet-502022.074979.794.4———————————————
EFToptimization-based=true, 2D ground truth provided=true2023.1249.3—————————————————
BOACD (Cross-domain evaluation)=true, V (Video-based method)=false2023.0349.577.2————————————————
Hierarchical matrix-Fisher (Hierarchical)Number of Samples=10, Backbone=HRNet2021.1049.679————————————————
ROMPCategory=Single Stage, Backbone=ResNet-50, training_on_3DPW=true2022.0749.779.794.7———————————————
FS + WS + OPTRegime=video, Optimization=KA + BA + S, Number of frames=16 frames2020.0349.880.2————————————————