List your company

3D Human Pose Estimation on Human3.6M

16.5MPJPEPoseMoE
15.3822.9430.538.06Aug 26, 2021Jun 8, 2022Mar 22, 2023Jan 2, 2024Oct 15, 2024Jul 28, 2025May 11, 2026

Evaluation results

MethodLinks
PoseMoENetwork=Mixture-of-Experts, Seq2Seq=true, T=243, Parameter=17.9M, MACs=66.3G, MACs/frame=272M2025.1216.5—————————————————————————————————————————————————
Volumetric2D pose detector fine-tuned=true2023.1217.7—————————————————————————————————————————————————
KTPFormerVenue=CVPR’24, Network=Lifting-Based, Seq2Seq=true, T=243, Parameter=33.7M, MACs=69.5G, MACs/frame=286M2025.1219—————————————————————————————————————————————————
Algebraic Triangulation2D pose detector fine-tuned=true2023.1219.2—————————————————————————————————————————————————
GLA-GCNVenue=ICCV’23, Network=Lifting-Based, Seq2Seq=false, T=243, Parameter=1.3M, MACs=1.5G, MACs/frame=1556M2025.1221—————————————————————————————————————————————————
STCFormerVenue=CVPR’23, Network=Lifting-Based, Seq2Seq=true, T=243, Parameter=4.7M, MACs=19.6G, MACs/frame=80M2025.1221.3—————————————————————————————————————————————————
PoseRetNetVenue=ECCV’24, Network=Lifting-Based, Seq2Seq=true, T=243, Parameter=25.2M, MACs=104.5G, MACs/frame=430M2025.1221.5—————————————————————————————————————————————————
MixSTEVenue=CVPR’22, Network=Lifting-Based, Seq2Seq=true, T=243, Parameter=33.6M, MACs=139.0G, MACs/frame=572M2025.1221.6—————————————————————————————————————————————————
Geometry-Biased TransformerBackbone=ResNet-152, 2D pose detector fine-tuned=true2023.1222.7—————————————————————————————————————————————————
TransFusion2D pose detector fine-tuned=true2023.1225.8—————————————————————————————————————————————————
Geometry-Biased TransformerBackbone=HRNet, 2D pose detector fine-tuned=false2023.1226—————————————————————————————————————————————————
Cross View Fusion2D pose detector fine-tuned=true2023.1226.2—————————————————————————————————————————————————
Epipolar Transformers2D pose detector fine-tuned=true2023.1227.1—————————————————————————————————————————————————
HeatFormerOptimization iterations=iter3, Translation estimation protocol=ground-truth 3D joints2024.1228.6————————————————————23.1————————————————————————————
P-STMOVenue=ECCV’22, Network=Lifting-Based, Seq2Seq=false, T=243, Parameter=6.2M, MACs=0.7G, MACs/frame=740M2025.1229.3—————————————————————————————————————————————————
HeatFormerOptimization iterations=iter42024.1229.5————————————————————22.4————————————————————————————
HTP (Ours)Type=Diffusion, F=243, CE=false, Detector=CPN, MACs (G)=87.7, Params (M)=37.52025.0829.9—————————————————————————————————————————23.3———————
Remelli et al.2D pose detector fine-tuned=true2023.1230.2—————————————————————————————————————————————————
GraphMLPInput Source=GT, Sequence length (T)=2432022.0630.3—————————————————————————————————————————————————
MHFormerVenue=CVPR’22, Network=Lifting-Based, Seq2Seq=false, T=351, Parameter=30.9M, MACs=7.1G, MACs/frame=7096M2025.1230.5—————————————————————————————————————————————————
HeatFormerOptimization iterations=iter32024.1230.7————————————————————23.3————————————————————————————
U-HMR2024.1231————————————————————22.8————————————————————————————
PoseFormerInput Source=GT, Sequence length (T)=812022.0631.3—————————————————————————————————————————————————
FinePOSEType=Diffusion, Publication=CVPR’24, F=243, CE=false, Detector=CPN, MACs (G)=146, Params (M)=200.62025.0831.9—————————————————————————————————————————25———————
SRNetInput Source=GT, Sequence length (T)=2432022.0632—————————————————————————————————————————————————
Anatomy3DInput Source=GT, Sequence length (T)=2432022.0632.3—————————————————————————————————————————————————
PaFF2024.1233————————————————————26.9————————————————————————————
KTPFormerType=Diffusion, Publication=CVPR’24, F=243, CE=false, Detector=CPN, MACs (G)=139.1, Params (M)=36.32025.0833—————————————————————————————————————————26.2———————
BaselineInput Source=GT, Sequence length (T)=2432022.0635.2—————————————————————————————————————————————————
D3DPType=Diffusion, Publication=ICCV’23, F=243, CE=false, Detector=CPN, MACs (G)=139.1, Params (M)=34.82025.0835.4—————————————————————————————————————————28.7———————
HYPERPOSEGeometry=H^d, Temporal Receptive Field (T)=243, 2D Keypoints=CPN2026.0536—————————————————————————————————————————29.11——————35.08
MeTRAbs-ACAE-Lvariant=L2024.0736.5————————————————————27.8————————————————————————————
MambaposeTemplate-based=false, Kinematics-based=false, Online=false2026.0136.5—————————————————————————————————————————28.6———————
DiffPoseAcceleration strategy=w/o DDIM2022.1136.7—————————————————————————173———————————————————————
DiffPoseAcceleration strategy=w/ DDIM2022.1136.9—————————————————————————671———————————————————————
SAMA-LType=Mamba, Publication=ICCV’25, F=243, CE=false, Detector=CPN, MACs (G)=53.2, Params (M)=17.32025.0836.9—————————————————————————————————————————31.3———————
DiffposeType=Diffusion, Publication=CVPR’23, F=243, CE=false, Detector=CPN2025.0836.9—————————————————————————————————————————28.7———————
PoseMambaTemplate-based=false, Kinematics-based=false, Online=false2026.0137.1—————————————————————————————————————————31.5———————
PoseMamba-XType=Mamba, Publication=AAAI’25, F=243, CE=false, Detector=CPN, MACs (G)=109.9, Params (M)=26.52025.0837.1—————————————————————————————————————————31.5———————
ST-GCNInput Source=GT, Sequence length (T)=92022.0637.2—————————————————————————————————————————————————
MotionBERTType=Transformer, Publication=ICCV’23, F=243, CE=false, Detector=SH, Setting=Finetune Setting, MACs (G)=174.8, Params (M)=42.52025.0837.5—————————————————————————————————————————————————
TCNInput Source=GT, Sequence length (T)=2432022.0637.8—————————————————————————————————————————————————
TCPFormerSeq2Seq=true, Number of input frames=243, Parameter=35.1M, MACs=109.2G, MACs/frames=449M2025.0137.9————————————————————31.7————————————————————————————
MotionAGFormerGeometry=Euc., Temporal Receptive Field (T)=243, 2D Keypoints=CPN2026.0538.4—————————————————————————————————————————32.4——————38.16
HTP w/ MotionBERTType=Transformer, F=243, CE=false, Detector=SH, Setting=Scratch Setting, MACs (G)=101.7, Params (M)=47.62025.0838.9—————————————————————————————————————————————————
MotionBERTSeq2Seq=true, Number of input frames=243, Parameter=42.3M, MACs=174.8G, MACs/frames=719M2025.0139.2————————————————————32.9————————————————————————————
MotionBERTType=Transformer, Publication=ICCV’23, F=243, CE=false, Detector=SH, Setting=Scratch Setting, MACs (G)=174.8, Params (M)=42.52025.0839.2—————————————————————————————————————————————————
MotionBERTGeometry=Euc., Temporal Receptive Field (T)=243, 2D Keypoints=CPN2026.0539.2—————————————————————————————————————————32.9——————39
NLF-Lvariant=L2024.0739.7————————————————————28.5————————————————————————————
CA-PF-HRNet-482D Pose Detector=HRNet-48 [53], Frame=1, FLOPs (G) for the Lifting Module=0.62023.1139.8————————————————————32.7————————————————————————————
TC-MixSTEType=Transformer, Publication=TMM’24, F=243, CE=false, Detector=CPN2025.0839.9—————————————————————————————————————————31.9———————
HTP w/ MixSTEType=Transformer, F=243, CE=false, Detector=CPN, MACs (G)=87.6, Params (M)=36.42025.0839.9—————————————————————————————————————————29.9———————
KTPFormerSeq2Seq=true, Number of input frames=243, Parameter=33.7M, MACs=69.5G, MACs/frames=286M2025.0140.1————————————————————31.9————————————————————————————
KTPFormerTemplate-based=false, Kinematics-based=false, Online=false2026.0140.1—————————————————————————————————————————31.9———————
MeTRAbs-ACAE-Svariant=S2024.0740.2————————————————————31.1————————————————————————————
PoseAnchorTemplate-based=false, Kinematics-based=false, Online=false2026.0140.3—————————————————————————————————————————32.1———————
NLF-Svariant=S2024.0740.4————————————————————30.6————————————————————————————
TCPFormerSeq2Seq=true, Number of input frames=81, Parameter=35.0M, MACs=36.4G, MACs/frames=150M2025.0140.5————————————————————33.7————————————————————————————
STCFormerType=Transformer, Publication=CVPR’23, F=243, CE=false, Detector=CPN, MACs (G)=78.2, Params (M)=18.92025.0840.5—————————————————————————————————————————31.8———————
STCFormerGeometry=Euc., Temporal Receptive Field (T)=243, 2D Keypoints=CPN2026.0540.5—————————————————————————————————————————31.8———————
Zhang2022.1140.9—————————————————————————974———————————————————————
MixSTEVenue=CVPR'22, 2D Pose Detector=CPN [8], Frame=243, FLOPs (G) for the Lifting Module=277.42023.1140.9————————————————————32.6————————————————————————————
MixSTESeq2Seq=true, Number of input frames=243, Parameter=33.6M, MACs=139.0G, MACs/frames=572M2025.0140.9————————————————————32.6————————————————————————————
MixSTEType=Transformer, Publication=CVPR’22, F=243, CE=false, Detector=CPN, MACs (G)=139.0, Params (M)=33.82025.0840.9—————————————————————————————————————————30.6———————
MixSTEGeometry=Euc., Temporal Receptive Field (T)=243, 2D Keypoints=CPN2026.0540.9—————————————————————————————————————————32.6———————
STCFormerSeq2Seq=true, Number of input frames=243, Parameter=4.7M, MACs=19.6G, MACs/frames=80M2025.0141————————————————————32————————————————————————————
HOTType=Transformer, Publication=CVPR’24, F=243, CE=false, Detector=CPN, MACs (G)=83.8, Params (M)=35.02025.0841—————————————————————————————————————————————————
CA-PF-HRNet-322D Pose Detector=HRNet-32 [53], Frame=1, FLOPs (G) for the Lifting Module=0.62023.1141.4————————————————————33.5————————————————————————————
CA-PF-CPN2D Pose Detector=CPN [8], Frame=1, FLOPs (G) for the Lifting Module=0.62023.1141.6————————————————————33.9————————————————————————————
Tang et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false2025.0642——————————————————————————40.64338.340.243.552.640.340.151.857.742.839.842.32829.5————————
Zhou et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false2025.0642——————————————————————————40.242.738.640.842.65040.340.252.555.143.641.342.929.529.5————————
ProHMROutput type=Minimum error of distribution2021.0842.4————————————————————32.9————————————————————————————
Li and LeeOutput type=Minimum error of distribution2021.0842.6————————————————————34.4————————————————————————————
P-STMOSeq2Seq=false, Number of input frames=243, Parameter=6.2M, MACs=0.7G, MACs/frames=740M2025.0142.8————————————————————34.4————————————————————————————
P-STMOType=Transformer, Publication=ECCV’22, F=243, CE=true, Detector=CPN, MACs (G)=1.7, Params (M)=7.02025.0842.8—————————————————————————————————————————34.4———————
DualFormerType=Transformer, Publication=TCSVT’24, F=351, CE=false, Detector=CPN2025.0842.8—————————————————————————————————————————34.4———————
DiffposeType=Diffusion, Publication=ICCV’23, F=64, CE=false2025.0842.9—————————————————————————————————————————30.8———————
Li2022.1143—————————————————————————328———————————————————————
MHFormerVenue=CVPR'22, 2D Pose Detector=CPN [8], Frame=351, FLOPs (G) for the Lifting Module=14.22023.1143————————————————————34.4————————————————————————————
P-STMOVenue=ECCV'22, 2D Pose Detector=CPN [8], Frame=243, FLOPs (G) for the Lifting Module=1.52023.1143————————————————————34.4————————————————————————————
MHFormerSeq2Seq=false, Number of input frames=351, Parameter=30.9M, MACs=7.1G, MACs/frames=7096M2025.0143————————————————————34.4————————————————————————————
Li et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false2025.0643——————————————————————————39.243.140.140.944.951.240.641.353.560.343.741.143.829.830.6————————
MHFormerType=Transformer, Publication=CVPR’22, F=351, CE=true, Detector=CPN, MACs (G)=9.6, Params (M)=24.72025.0843—————————————————————————————————————————34.4———————
MHFormerGeometry=Euc., Temporal Receptive Field (T)=351, 2D Keypoints=CPN2026.0543—————————————————————————————————————————————————
StridedTrans.Venue=TMM'22, 2D Pose Detector=CPN [8], Frame=351, FLOPs (G) for the Lifting Module=2.12023.1143.7————————————————————35.2————————————————————————————
Li et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false, Source=TMM'222025.0643.7——————————————————————————40.343.340.242.345.652.341.840.555.960.644.24344.23030.2————————
CLIFF + PROMorthogonalization=PROM2023.1243.8————————————————————30.4————————————————————————————
GraphMLPInput Source=CPN, Sequence length (T)=2432022.0643.8—————————————————————————————————————————————————
SemGCNGeometry=Euc., Temporal Receptive Field (T)=1, 2D Keypoints=CPN2026.0543.8—————————————————————————————————————————————————
Anatomy3DInput Source=CPN, Sequence length (T)=2432022.0644.1—————————————————————————————————————————————————
Algebraic TriangulationBackbone=HRNet, 2D pose detector fine-tuned=false2023.1244.2—————————————————————————————————————————————————
PoseFormerVenue=ICCV'21, 2D Pose Detector=CPN [8], Frame=81, FLOPs (G) for the Lifting Module=1.42023.1144.3————————————————————34.6————————————————————————————
PoseFormerInput Source=CPN, Sequence length (T)=812022.0644.3—————————————————————————————————————————————————
Zheng et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false2025.0644.3——————————————————————————41.544.839.842.546.551.642.14253.360.745.543.346.131.832.2————————
PoseFormerType=Transformer, Publication=ICCV’21, F=81, CE=true, Detector=CPN, MACs (G)=1.6, Params (M)=9.62025.0844.3—————————————————————————————————————————36.5———————
PoseFormerGeometry=Euc., Temporal Receptive Field (T)=81, 2D Keypoints=CPN2026.0544.3—————————————————————————————————————————34.6———————
GLA-GCNSeq2Seq=false, Number of input frames=243, Parameter=1.3M, MACs=1.5G, MACs/frames=1556M2025.0144.4————————————————————34.8————————————————————————————
GLA-GCNType=GCN, Publication=ICCV’23, F=243, CE=true, Detector=CPN2025.0844.4—————————————————————————————————————————34.8———————
PostureHMR2024.1244.5————————————————————31————————————————————————————
Wang et al.Temporal Information=true, 2D Keypoint Input=CPN detections, Pose Refinement=false2025.0644.5——————————————————————————40.242.542.641.146.756.741.442.356.260.446.342.246.231.731————————