Multi-person Pose Estimation on CrowdPose (test)
83.8APRTMO-l
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTMO-lMethod Category=One-stage, #Params=44.8M, Additional training data=true2023.12 | 83.8 | — | — | 88.8 | 84.7 | 77.2 | — | — | — | — | — | — | — | — | — | — | |
| BUCTD-COAM-W48++σInput size=384×288, Conditional Input=PETR, Generative Sampling=true, Category=Hybrid2023.06 | 78.5 | — | — | — | — | — | 83.9 | 79 | 72.3 | — | — | — | — | — | — | — | |
| I2R-NetMethod architecture=Top-down, Input resolution=256x192, 1st stage backbone=HRFormer-B2022.06 | 77.4 | 93.6 | 83.3 | — | 78.1 | — | 83.8 | — | 69.3 | — | 94.5 | 85.5 | — | — | — | 80.3 | |
| PoseFix (HRNet-preNet-W48)++σInput size=384×288, Conditional Input=PETR, Generative Sampling=true, Category=Hybrid2023.06 | 76.8 | — | — | — | — | — | 82.3 | 77.4 | 70.2 | — | — | — | — | — | — | — | |
| BUCTD-COAM-W48++Input size=384×288, Conditional Input=PETR, Category=Hybrid2023.06 | 76.7 | — | — | — | — | — | 82.6 | 77.2 | 70.4 | — | — | — | — | — | — | — | |
| ED-PoseBackbone=Swin-L, Architecture Type=End-to-end, Additional training data beyond COCO=true2025.06 | 76.6 | 92.4 | 83.3 | 83 | 77.3 | 68.3 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseParadigm=One-Stage, Backbone=Swin-L, Loss=Regression, Flipping test=false, Pre-trained=Objects3652023.02 | 76.6 | 92.4 | 83.3 | 83 | 77.3 | 68.3 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseMethod Category=One-stage, #Params=218.0M, Additional training data=true2023.12 | 76.6 | — | — | 83.3 | 77.3 | 68.3 | — | — | — | — | — | — | — | — | — | — | |
| I2R-NetMethod architecture=Top-down, Input resolution=256x192, 1st stage backbone=TransPose-H2022.06 | 76.3 | 93.5 | 82.2 | — | 77 | — | 83.2 | — | 67.4 | — | 94 | 84.4 | — | — | — | 79.1 | |
| CIGPose-x#Params=50.4M, Input Resolution=384x288, Method Type=Top-down2026.03 | 75.8 | — | — | — | — | — | 84.2 | 77.3 | 63.6 | — | — | — | — | — | — | — | |
| BUCTD-COAM-W48+Input size=384×288, Conditional Input=CID-W32, Category=Hybrid2023.06 | 75.3 | — | — | — | — | — | 81.1 | 75.9 | 68.4 | — | — | — | — | — | — | — | |
| DETRPose-XBackbone=HGNetv2-B5, Architecture Type=End-to-end2025.06 | 75.1 | 92.1 | 81.3 | 81.3 | 75.7 | 68.1 | — | — | — | — | — | — | — | — | — | — | |
| CIGPose-l#Params=28.8M, Input Resolution=384x288, Method Type=Top-down2026.03 | 74.2 | — | — | — | — | — | 82.9 | 75.6 | 62.5 | — | — | — | — | — | — | — | |
| GroupPoseBackbone=Swin-L, Architecture Type=End-to-end2025.06 | 74.1 | 91.3 | 80.4 | 80.8 | 74.7 | 66.4 | — | — | — | — | — | — | — | — | — | — | |
| CIGPose-l#Params=28.4M, Input Resolution=256x192, Method Type=Top-down2026.03 | 73.7 | — | — | — | — | — | 82.8 | 75.1 | 61.2 | — | — | — | — | — | — | — | |
| DETRPose-LBackbone=HGNetv2-B4, Architecture Type=End-to-end2025.06 | 73.3 | 91.6 | 79.4 | 79.5 | 74 | 66.1 | — | — | — | — | — | — | — | — | — | — | |
| RTMO-LBackbone=CSPDarknet, Architecture Type=Non-end-to-end2025.06 | 73.2 | 90.7 | 79.3 | 79.2 | 74.1 | 65.3 | — | — | — | — | — | — | — | — | — | — | |
| RTMO-lMethod Category=One-stage, #Params=44.8M, Additional training data=false2023.12 | 73.2 | — | — | 79.2 | 74.1 | 65.3 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseBackbone=Swin-L, Architecture Type=End-to-end2025.06 | 73.1 | 90.5 | 79.8 | 80.5 | 73.8 | 63.8 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseParadigm=One-Stage, Backbone=Swin-L, Loss=Regression, Flipping test=false2023.02 | 73.1 | 90.5 | 79.8 | 80.5 | 73.8 | 63.8 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseMethod Category=One-stage, #Params=218.0M, Additional training data=false2023.12 | 73.1 | — | — | 80.5 | 73.8 | 63.8 | — | — | — | — | — | — | — | — | — | — | |
| BUCTD-COAM-W48Input size=384×288, Category=Hybrid2023.06 | 72.9 | — | — | — | — | — | 79.2 | 73.4 | 66.1 | — | — | — | — | — | — | — | |
| HRNetMethod architecture=Top-down, Input resolution=384x2882022.06 | 72.8 | 92.1 | 78.7 | — | 73.3 | — | 81.3 | — | 65.5 | — | — | — | — | — | — | — | |
| QueryPoseBackbone=Swin-L, Architecture Type=End-to-end2025.06 | 72.7 | 91.7 | 78.1 | 79.5 | 73.4 | 65.4 | — | — | — | — | — | — | — | — | — | — | |
| QueryPose-SwinLMethod Type=End-to-End MPPE, Backbone=Swin-Large, Multi-scale testing=false2022.12 | 72.7 | 91.7 | 78.1 | 79.5 | 73.4 | 65.4 | — | — | — | — | — | — | — | — | — | — | |
| ER-pose-lBackbone=CSPDarknet, Input size=640×640, Params(M)=39.952026.03 | 72.5 | 91.2 | 79.1 | 78.8 | 73.1 | 65 | — | — | — | — | — | — | — | — | — | — | |
| HRFormer-BMethod architecture=Top-down, Input resolution=256x1922022.06 | 72.4 | 91.5 | 77.9 | — | 73.5 | — | 80 | — | 62.4 | — | 92.7 | 81 | — | — | — | 75.6 | |
| HRFormer-BParadigm=Top-Down, Loss=Heatmap-based, Flipping test=false2023.02 | 72.4 | 91.5 | 77.9 | 80 | 73.5 | 62.4 | — | — | — | — | — | — | — | — | — | — | |
| HRFormer-BMethod Category=Top-down, #Params=43.2M, Additional training data=false2023.12 | 72.4 | — | — | 80 | 73.5 | 62.4 | — | — | — | — | — | — | — | — | — | — | |
| HRFormer-B#Params=43.2M, Input Resolution=256x192, Method Type=Top-down2026.03 | 72.4 | — | — | — | — | — | 80 | 73.5 | 62.4 | — | — | — | — | — | — | — | |
| CID-W48Input size=640, Category=Single-stage2023.06 | 72.3 | — | — | — | — | — | 78.7 | 73 | 64.8 | — | — | — | — | — | — | — | |
| I2R-NetMethod architecture=Top-down, Input resolution=256x192, Version=Vanilla version, 1st stage backbone=HRNet2022.06 | 72.3 | 92.4 | 77.9 | — | 73.2 | — | 79.9 | — | 62.8 | — | 93.2 | 81.9 | — | — | — | 76.5 | |
| CIDMethod Category=One-stage, #Params=65.4M, Additional training data=false2023.12 | 72.3 | — | — | 78.7 | 72.1 | 64.8 | — | — | — | — | — | — | — | — | — | — | |
| BAPoseInput size=512, Category=Bottom-Up2023.06 | 72.2 | — | — | — | — | — | 79.9 | 73.4 | 61.3 | — | — | — | — | — | — | — | |
| BAPose (W32)Input Size=512, Approach=Bottom-Up, Testing Mode=Single-Scale Testing2021.12 | 72.2 | 89.6 | 78 | 79.9 | 73.4 | 61.3 | — | — | — | — | — | — | — | — | — | — | |
| QueryPose-W48Method Type=End-to-End MPPE, Backbone=HRNet-W48, Multi-scale testing=false2022.12 | 72.1 | 89.4 | 78 | 79.8 | 73.3 | 63 | — | — | — | — | — | — | — | — | — | — | |
| DETRPose-MBackbone=HGNetv2-B2, Architecture Type=End-to-end2025.06 | 72 | 91 | 77.8 | 78.6 | 72.6 | 64.5 | — | — | — | — | — | — | — | — | — | — | |
| PETRInput size=800, Category=Single-stage2023.06 | 72 | — | — | — | — | — | 78 | 72.5 | 65.4 | — | — | — | — | — | — | — | |
| TransPose-HMethod architecture=Top-down, Input resolution=256x1922022.06 | 71.8 | 91.5 | 77.8 | — | 72.9 | — | 79.5 | — | 62.2 | — | 92.7 | 80.4 | — | — | — | 75.2 | |
| TransPose-HParadigm=Top-Down, Loss=Heatmap-based, Flipping test=false2023.02 | 71.8 | 91.5 | 77.8 | 79.5 | 72.9 | 62.2 | — | — | — | — | — | — | — | — | — | — | |
| TransPose-HMethod Category=Top-down, Additional training data=false2023.12 | 71.8 | — | — | 79.5 | 72.9 | 62.2 | — | — | — | — | — | — | — | — | — | — | |
| TransPose-H#Params=-, Input Resolution=256x192, Method Type=Top-down2026.03 | 71.8 | — | — | — | — | — | 79.5 | 72.9 | 62.2 | — | — | — | — | — | — | — | |
| PETRBackbone=Swin-L, Architecture Type=End-to-end2025.06 | 71.6 | 90.4 | 78.3 | 77.3 | 72 | 65.8 | — | — | — | — | — | — | — | — | — | — | |
| SWAHR-W48Method Type=End-to-End MPPE, Backbone=HRNet-W48, Multi-scale testing=false2022.12 | 71.6 | 88.5 | 77.6 | 78.9 | 72.4 | 63 | — | — | — | — | — | — | — | — | — | — | |
| SWAHRParadigm=Bottom-Up, Backbone=HRNet-w32, Loss=Heatmap-based, Flipping test=true2023.02 | 71.6 | 88.5 | 77.6 | 78.9 | 72.4 | 63 | — | — | — | — | — | — | — | — | — | — | |
| PETRParadigm=One-Stage, Backbone=Swin-L, Loss=Regression + Heatmap, Flipping test=false2023.02 | 71.6 | 90.4 | 78.3 | 77.3 | 72 | 65.8 | — | — | — | — | — | — | — | — | — | — | |
| SWAHRMethod Category=Bottom-up, #Params=63.8M, Additional training data=false2023.12 | 71.6 | — | — | 78.9 | 72.4 | 63 | — | — | — | — | — | — | — | — | — | — | |
| PETRMethod Category=One-stage, #Params=220.5M, Additional training data=false2023.12 | 71.6 | — | — | 77.3 | 72 | 65.8 | — | — | — | — | — | — | — | — | — | — | |
| SWAHR#Params=63.8M, Input Resolution=256x192, Method Type=Bottom-up2026.03 | 71.6 | — | — | — | — | — | 78.9 | 72.4 | 63 | — | — | — | — | — | — | — | |
| CIGPose-m#Params=14.4M, Input Resolution=256x192, Method Type=Top-down2026.03 | 71.4 | — | — | — | — | — | 81 | 72.7 | 58.9 | — | — | — | — | — | — | — | |
| CID-W32Input size=512, Category=Single-stage2023.06 | 71.3 | — | — | — | — | — | 77.4 | 72.1 | 63.9 | — | — | — | — | — | — | — | |
| HRNetMethod architecture=Top-down, Input resolution=256x1922022.06 | 71.3 | 91.1 | 77.5 | — | 71.4 | — | 80.5 | — | 62.5 | — | — | — | — | — | — | — | |
| HRNetParadigm=Top-Down, Backbone=HRNet-w48, Loss=Heatmap-based, Flipping test=true2023.02 | 71.3 | 91.1 | 77.5 | 80.5 | 71.4 | 62.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNetMethod Category=Top-down, #Params=28.5M, Additional training data=false2023.12 | 71.3 | — | — | 80.5 | 71.4 | 62.5 | — | — | — | — | — | — | — | — | — | — | |
| HRNet#Params=28.5M, Input Resolution=256x192, Method Type=Top-down2026.03 | 71.3 | — | — | — | — | — | 80.5 | 71.4 | 62.5 | — | — | — | — | — | — | — | |
| CIDBackbone=HRNet-W32, Input size=512×512, Params(M)=29.42026.03 | 71.2 | 89.8 | 76.7 | 77.9 | 71.9 | 63.8 | — | — | — | — | — | — | — | — | — | — | |
| RTMO-MBackbone=CSPDarknet, Architecture Type=Non-end-to-end2025.06 | 71.1 | 87.7 | 77.1 | 77.4 | 71.9 | 63.4 | — | — | — | — | — | — | — | — | — | — | |
| RTMO-mMethod Category=One-stage, #Params=22.6M, Additional training data=false2023.12 | 71.1 | — | — | 77.4 | 71.9 | 63.4 | — | — | — | — | — | — | — | — | — | — | |
| OPEC-NetBackbone=ResNet101, Detector=YOLO-v3, Paradigm=Top-Down2021.01 | 70.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OPEC-NetMethod architecture=Top-down, Input resolution=256x1922022.06 | 70.6 | 86.8 | 75.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RTMPose-mMethod Category=Top-down, #Params=13.5M, Additional training data=false2023.12 | 70.6 | — | — | 79.9 | 71.9 | 58.2 | — | — | — | — | — | — | — | — | — | — | |
| RTMPose-m#Params=13.5M, Input Resolution=256x192, Method Type=Top-down2026.03 | 70.6 | — | — | — | — | — | 79.9 | 71.9 | 58.2 | — | — | — | — | — | — | — | |
| MIPNet-W48Input size=384×288, Category=Hybrid2023.06 | 70 | — | — | — | — | — | 78.1 | 71.1 | 59.4 | — | — | — | — | — | — | — | |
| CenterGroupcategory=Bottom-up, backbone=HrHRNet-W48, multi-scale test=true2021.10 | 70 | 88.9 | 75.1 | 76.8 | 70.7 | 62.2 | — | — | — | — | — | — | — | — | — | — | |
| MIPNetBackbone=HRNet-W48-384, Detector=Faster R-CNN, Paradigm=Top-Down2021.01 | 70 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIPNetInput Size=512, Approach=Top-Down, Testing Mode=Single-Scale Testing2021.12 | 70 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIPNetstage=two-stage, TTA=true2021.11 | 70 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterGroup-W48Latency=1060ms, stage=single-stage, TTA=true2021.11 | 70 | 88.9 | 75.7 | 77.3 | 70.8 | 63.2 | — | — | — | — | — | — | — | — | — | — | |
| CenterGroup-W48Method Type=End-to-End MPPE, Backbone=HRNet-W48, Multi-scale testing=true2022.12 | 70 | 88.9 | 75.1 | 76.8 | 70.7 | 62.2 | — | — | — | — | — | — | — | — | — | — | |
| ED-PoseBackbone=ResNet-50, Architecture Type=End-to-end2025.06 | 69.9 | 88.6 | 75.8 | 77.7 | 70.6 | 60.9 | — | — | — | — | — | — | — | — | — | — | |
| PINet Multi-scaleMethod architecture=Bottom-up, Multi-scale evaluation=true2022.06 | 69.9 | 89.1 | 75.6 | — | 70.5 | — | 76.4 | — | 62.2 | — | — | — | — | — | — | — | |
| ED-PoseParadigm=One-Stage, Backbone=ResNet-50, Loss=Regression, Flipping test=false2023.02 | 69.9 | 88.6 | 75.8 | 77.7 | 70.6 | 60.9 | — | — | — | — | — | — | — | — | — | — | |
| AIDBackbone=HigherHrnet-W48, Input size=640 x 640, Multi-scale testing=true, Baseline=UDPv12020.08 | 69.7 | 88.4 | 74.9 | — | 77.9 | — | — | — | — | — | — | — | — | 70.3 | 61.4 | — | |
| CenterGroupInput size=640, Category=Bottom-Up2023.06 | 69.4 | — | — | — | — | — | 76.6 | 70 | 61.5 | — | — | — | — | — | — | — | |
| ER-pose-mBackbone=CSPDarknet, Input size=640×640, Params(M)=23.362026.03 | 69.4 | 90.1 | 75.9 | 76.4 | 69.9 | 61.9 | — | — | — | — | — | — | — | — | — | — | |
| HRNet-W48Input size=384×288, Category=Top-Down2023.06 | 69.3 | — | — | — | — | — | 77.7 | 70.6 | 57.8 | — | — | — | — | — | — | — | |
| HRNetBackbone=HRNet-W48-384, Detector=Faster R-CNN, Paradigm=Top-Down2021.01 | 69.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| YOLO-Pose-lBackbone=CSPDarknet, Input size=640×640, Params(M)=44.42026.03 | 69.3 | 90.6 | 76.2 | 76.1 | 70.2 | 61.5 | — | — | — | — | — | — | — | — | — | — | |
| AdaptivePose-W48Method Type=End-to-End MPPE, Backbone=HRNet-W48, Multi-scale testing=true2022.12 | 69.2 | 87.3 | 75 | 76.7 | 70 | 60.9 | — | — | — | — | — | — | — | — | — | — | |
| HigherHRNet + UDPBackbone=HRNet-W48, Input size=640 x 640, Multi-scale test=true, Trained with official implementation=true2019.11 | 69 | 88.5 | 74 | 76.9 | — | — | — | — | — | — | — | — | — | 69.5 | 60.7 | — | |
| PINetMethod architecture=Bottom-up2022.06 | 68.9 | 88.7 | 74.7 | — | 69.6 | — | 75.4 | — | 61.5 | — | — | — | — | — | — | — | |
| KAPAO-LLatency=215ms, stage=single-stage, TTA=true2021.11 | 68.9 | 89.4 | 75.6 | 76.6 | 69.9 | 59.5 | — | — | — | — | — | — | — | — | — | — | |
| KAPAO-lMethod Category=One-stage, #Params=77.0M, Additional training data=false2023.12 | 68.9 | — | — | 76.6 | 69.9 | 59.5 | — | — | — | — | — | — | — | — | — | — | |
| KAPAO-lBackbone=CSPDarknet, Input size=1280×1280, Params(M)=77.02026.03 | 68.9 | 89.4 | 75.6 | 76.6 | 69.9 | 59.5 | — | — | — | — | — | — | — | — | — | — | |
| PETRBackbone=ResNet-50, Architecture Type=End-to-end2025.06 | 68.8 | 87.5 | 76.3 | 62.7 | 77.7 | — | — | — | — | — | — | — | — | — | — | — | |
| HigherHRNetBackbone=HRNet-W48, Input size=640 x 640, Multi-scale test=true, Trained with official implementation=true2019.11 | 68.8 | 88.3 | 73.9 | 76.5 | — | — | — | — | — | — | — | — | — | 69.5 | 60.2 | — | |
| AlphaPose+Backbone=ResNet101, Detector=YOLO-v3, Paradigm=Top-Down2021.01 | 68.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AIDBackbone=HigherHrnet-W48, Input size=640 x 640, Multi-scale testing=false, Baseline=UDPv12020.08 | 68.2 | 87.3 | 73.6 | — | 75.4 | — | — | — | — | — | — | — | — | 68.6 | 60.4 | — | |
| UDPv1Backbone=HigherHrnet-W48, Input size=640 x 640, Multi-scale testing=true2020.08 | 68.2 | 88 | 72.9 | — | 76.6 | — | — | — | — | — | — | — | — | 68.7 | 59.9 | — | |
| HigherHRNet + UDPBackbone=HigherHRNet-W48, Input size=640 x 640, Multi-scale test=true, Trained with official implementation=false2019.11 | 68.2 | 88 | 72.9 | 76.6 | — | — | — | — | — | — | — | — | — | 68.7 | 59.9 | — | |
| HghrHRNet+UDPParadigm=Bottom-Up, Multi-scale=x2, x1, x0.52021.01 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MIPNetBackbone=ResNet101, Detector=YOLO-v3, Paradigm=Top-Down2021.01 | 68.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DEKRInput size=640, Category=Single-stage2023.06 | 68 | — | — | — | — | — | 76.6 | 68.8 | 58.4 | — | — | — | — | — | — | — | |
| DEKR (HRNet-W48)Backbone=HRNet-W48, Input size=640, Testing scale=multi-scale2021.04 | 68 | 85.5 | 73.4 | 76.6 | 68.8 | 58.4 | — | — | — | — | — | — | — | — | — | — | |
| DEKR-W48Latency=1170ms, stage=single-stage, TTA=true, Rescoring network=model-agnostic rescoring network2021.11 | 68 | 85.5 | 73.4 | 76.6 | 68.8 | 58.4 | — | — | — | — | — | — | — | — | — | — | |
| DEKR-W48Method Type=End-to-End MPPE, Backbone=HRNet-W48, Multi-scale testing=false2022.12 | 68 | 85.5 | 73.4 | 76.6 | 68.8 | 58.4 | — | — | — | — | — | — | — | — | — | — | |
| HigherHRNet + UDPBackbone=HRNet-W32, Input size=512 x 512, Multi-scale test=true, Trained with official implementation=true2019.11 | 67.8 | 88 | 72.7 | 76.4 | — | — | — | — | — | — | — | — | — | 68.3 | 59.3 | — | |
| HigherHRNet-W48Input size=640, Category=Bottom-Up2023.06 | 67.6 | — | — | — | — | — | 75.8 | 68.1 | 58.9 | — | — | — | — | — | — | — | |
| HigherHRNet-W48+Method category=Bottom-up, Multi-scale test=true, Backbone=HRNet-W482019.08 | 67.6 | 87.4 | 72.6 | 75.8 | 68.1 | 58.9 | — | — | — | — | — | — | — | — | — | — | |
| HigherHRNetBackbone=HigherHrnet-W48, Input size=640 x 640, Multi-scale testing=true2020.08 | 67.6 | 87.4 | 72.6 | — | 75.8 | — | — | — | — | — | — | — | — | 68.1 | 58.9 | — |