Human Pose Estimation on MPII (test)
97.8Shoulder PCKPCT (Swin-Large)
Evaluation results
| Method | Links | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PCT (Swin-Large)backbone=Swin-Large2023.03 | 97.8 | 98.9 | 94.8 | 91.1 | 93.6 | 93 | 89.7 | — | — | — | — | 94.3 | — | — | — | — | — | — | — | — | — | |
| Bin et al.larger image size=true2023.03 | 97.6 | 98.9 | 94.6 | 91.2 | 93.1 | 92.7 | 89.1 | — | — | — | — | 94.1 | — | — | — | — | — | — | — | — | — | |
| Ours*pre-training=HSSK dataset2020.02 | 97.5 | 98.8 | 94.4 | 91.2 | 93.2 | 92.2 | 89.3 | — | — | — | — | 94.1 | — | — | — | — | — | — | — | — | — | |
| OursBackbone=HRNetW32, Input Size=256 x 2562023.03 | 97.5 | 98.7 | 94 | 90.6 | 92.5 | 91.1 | 87.1 | — | — | — | — | 93.3 | — | — | — | — | — | — | — | — | — | |
| Bulat et al.extra training datasets=true2023.03 | 97.5 | 98.8 | 94.4 | 91.2 | 93.2 | 92.2 | 89.3 | — | — | — | — | 94.1 | — | — | — | — | — | — | — | — | — | |
| PCT (Swin-Base)backbone=Swin-Base2023.03 | 97.5 | 98.7 | 94.2 | 90.6 | 92.9 | 92.1 | 88.7 | — | — | — | — | 93.8 | — | — | — | — | — | — | — | — | — | |
| 4xRSN-50Backbone=RSN-50, Number of stages=42020.03 | 97.3 | 98.5 | 93.9 | 89.9 | 92 | 90.6 | 86.8 | — | — | — | — | 93 | — | — | — | — | — | — | — | — | — | |
| Xie et al.2023.03 | 97.3 | 98.7 | 93.7 | 90.2 | 92 | 90.3 | 86.5 | — | — | — | — | 93 | — | — | — | — | — | — | — | — | — | |
| Cai et al.2023.03 | 97.3 | 98.5 | 93.9 | 89.9 | 92 | 90.6 | 86.8 | — | — | — | — | 93 | — | — | — | — | — | — | — | — | — | |
| MSPN2019.01 | 97.1 | 98.4 | 93.2 | 89.2 | 92 | 90.1 | 85.5 | — | — | — | — | 92.6 | — | — | — | — | — | — | — | — | — | |
| Hg-oursBackbone=Hourglass, Components=CPF + PGNN, Training Data=MPII + LSP, Scale=five-scale input, Augmentation=horizontal flip testing2019.01 | 97 | 98.6 | 92.8 | 88.8 | 91.7 | 89.8 | 86.6 | — | — | — | — | 92.5 | — | — | — | — | — | — | — | — | — | |
| Zhang et al.2019.01 | 97 | 98.6 | 92.8 | 88.8 | 91.7 | 89.8 | 86.6 | — | — | — | — | 92.5 | — | — | — | — | — | — | — | — | — | |
| Ours2020.02 | 97 | 98.6 | 93 | 89.2 | 91.7 | 88.9 | 86 | — | — | — | — | 92.4 | — | — | — | — | — | — | — | — | — | |
| Zhang et al.Category=Heatmap-based2021.03 | 97 | 98.6 | 92.8 | 88.8 | 91.7 | 89.8 | 86.6 | — | — | — | — | 92.5 | — | — | — | — | — | — | — | — | — | |
| Zhang et al., arXiv'19Parameters=24M2020.04 | 97 | 98.6 | 92.8 | 88.8 | 91.7 | 89.8 | 86.6 | — | — | — | — | 92.5 | — | — | — | — | — | — | — | — | — | |
| Zhang et al.2023.03 | 97 | 98.6 | 92.8 | 88.8 | 91.7 | 89.8 | 86.6 | — | — | — | — | 92.5 | — | — | — | — | — | — | — | — | — | |
| Tang et al. [62]2019.02 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| HRNet-W32Input size=256x2562019.02 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Nie et al.# Param=26M, Deployment Cost=63G2018.11 | 96.9 | 98.6 | 93 | 89.1 | 91.7 | 89 | 86.2 | — | — | — | — | 92.4 | 65.9 | — | — | — | — | — | — | — | — | |
| Tang et al.citation=[42]2019.01 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al. (HRNet)2019.01 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al.2020.02 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al.2020.02 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al.2020.03 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al.2020.03 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al.#Params=15.5M, #FLOPs=33.6G2019.09 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al.Category=Heatmap-based2021.03 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al., ECCV'18Parameters=16M2020.04 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al., CVPR'19Parameters=29M2020.04 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al.Backbone=HRNetW322023.03 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Tang et al.2023.03 | 96.9 | 98.4 | 92.6 | 88.7 | 91.8 | 89.4 | 86.2 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Sun et al.2023.03 | 96.9 | 98.6 | 92.8 | 89 | 91.5 | 89 | 85.7 | — | — | — | — | 92.3 | — | — | — | — | — | — | — | — | — | |
| Chou et al.Method category=Detection based, citation=[11]2017.10 | 96.8 | 98.2 | 92.2 | 88 | 91.3 | 89.1 | 84.9 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Chou et al.2019.01 | 96.8 | 98.2 | 92.2 | 88 | 91.3 | 89.1 | 84.9 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Chou et al.2019.02 | 96.8 | 98.2 | 92.2 | 88 | 91.3 | 89.1 | 84.9 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Ke et al.2019.02 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.3 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Chou et al.2019.01 | 96.8 | 98.2 | 92.2 | 88 | 91.3 | 89.1 | 84.9 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Ke et al.2019.01 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.3 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Chou et al.2020.02 | 96.8 | 98.2 | 92.2 | 88 | 91.3 | 89.1 | 84.9 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Ke et al.2020.02 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.4 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Ke et al.2020.03 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.3 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Ke et al.Category=Heatmap-based2021.03 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.3 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Ke et al.2023.03 | 96.8 | 98.5 | 92.7 | 88.4 | 90.6 | 89.4 | 86.3 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| Yang et al.test time data augmentations=disabled2018.01 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | 92 | — | — | — | — | — | — | — | — | — | — | — | — | |
| PyraNet (Ours-B)Training data used=all the MPII training set2017.08 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al.2019.01 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al.2019.02 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al.# Param=28M, Deployment Cost=46G2018.11 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | 64.2 | — | — | — | — | — | — | — | — | |
| Yang et al.2019.01 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al.2020.02 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al.2020.03 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Yang et al., ICCV'17Parameters=27M2020.04 | 96.7 | 98.5 | 92.5 | 88.7 | 91.1 | 88.6 | 86 | — | — | — | — | 92 | — | — | — | — | — | — | — | — | — | |
| Adversarial Data AugmentationAdversarial Scaling and Rotating=true, Adversarial Hierarchical Occluding=true, Base model=Stacked HGs(8)2018.05 | 96.6 | 98.1 | 92.5 | 88.4 | 90.7 | 87.7 | 83.5 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Soft-argmax (Our method)Method category=Regression based2017.10 | 96.6 | 98.1 | 92 | 87.5 | 90.6 | 88 | 82.7 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Luvizon et al.2019.02 | 96.6 | 98.1 | 92 | 87.5 | 90.6 | 88 | 82.7 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| SimpleBaselineBackbone=ResNet-152, Input size=256x2562019.02 | 96.6 | 98.5 | 91.9 | 87.6 | 91.1 | 88.1 | 84.1 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Peng et al.# Param=26M, Deployment Cost=55G2018.11 | 96.6 | 98.1 | 92.5 | 88.4 | 90.7 | 87.7 | 83.5 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Luvizon et al.2019.01 | 96.6 | 98.1 | 92 | 87.5 | 90.6 | 88 | 82.7 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Ryou et al.2020.02 | 96.6 | 98.6 | 92.3 | 87.8 | 90.8 | 88.8 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Xiao et al.#Params=68.6M, #FLOPs=20.9G2019.09 | 96.6 | 98.5 | 91.9 | 87.6 | 91.1 | 88.1 | 84.1 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Xiao et al.2023.03 | 96.6 | 98.5 | 91.9 | 87.6 | 91.1 | 88.1 | 84.1 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Xiao et al.2023.03 | 96.6 | 98.5 | 91.9 | 87.6 | 91.1 | 88.1 | 84.1 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| PyraNet (Ours-A)Training data used=training set used in [51]2017.08 | 96.5 | 98.4 | 91.9 | 88.2 | 91.1 | 88.6 | 85.3 | — | — | — | — | 91.8 | — | — | — | — | — | — | — | — | — | |
| Chen et al.Method category=Detection based, citation=[8]2017.10 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al.2019.01 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al.2019.02 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al.2019.01 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al. ICCV'172020.02 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al. TPAMI'192020.02 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Chen et al.2020.03 | 96.5 | 98.1 | 92.5 | 88.5 | 90.2 | 89.6 | 86 | — | — | — | — | 91.9 | — | — | — | — | — | — | — | — | — | |
| Adversarial PoseNetmulti-task=true, discriminators=true, stacked layers=42017.04 | 96.4 | 98.6 | 92.4 | 88.6 | 91.5 | 88.6 | 85.7 | — | — | — | — | 92.1 | — | — | — | — | — | — | — | — | — | |
| ResNet-oursBackbone=modified ResNet-50, Components=CPF + PGNN2019.01 | 96.4 | 98.2 | 91.6 | 87.1 | 91.2 | 88 | 83.6 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Tang et al.2019.02 | 96.4 | 97.4 | 92.1 | 87.7 | 90.2 | 87.7 | 84.3 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| FPD# Param=3M, Deployment Cost=9G2018.11 | 96.4 | 98.3 | 91.5 | 87.4 | 90.9 | 87.1 | 83.7 | — | — | — | — | 91.1 | 63.5 | — | — | — | — | — | — | — | — | |
| Tang et al.citation=[44]2019.01 | 96.4 | 97.4 | 92.1 | 87.7 | 90.2 | 87.7 | 84.3 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Ours (small)model_size=small2020.02 | 96.4 | 98.5 | 91.5 | 87.2 | 90.7 | 86.9 | 83.6 | — | — | — | — | 91.1 | — | — | — | — | — | — | — | — | — | |
| FPD (Knowledge Distillation)#Params=3M, #FLOPs=9G2019.09 | 96.4 | 98.3 | 91.5 | 87.4 | 90.9 | 87.1 | 83.7 | — | — | — | — | 91.1 | — | — | — | — | — | — | — | — | — | |
| Zhang et al., CVPR'19Parameters=3M2020.04 | 96.4 | 98.3 | 91.5 | 87.4 | 90.9 | 87.1 | 83.7 | — | — | — | — | 91.1 | — | — | — | — | — | — | — | — | — | |
| Bulat et al., FG'20Parameters=9M2020.04 | 96.4 | 98.5 | 91.5 | 87.2 | 90.7 | 86.9 | 83.6 | — | — | — | — | 91.1 | — | — | — | — | — | — | — | — | — | |
| Newell et al.test time data augmentations=disabled2018.01 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | 90.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Chu et al.2018.05 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Stacked HGs(8)number of stacks=82018.05 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Newell et al.2017.02 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Multi-Context Attention (Ours)2017.02 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Stacked Hourglass2017.04 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| GNetflip_augmentation=true, scales=1.0, 0.75, post-processing=cross-heatmap NMS2017.05 | 96.3 | 98.1 | 92.2 | 87.8 | 90.6 | 87.6 | 82.7 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Newell et al.2017.05 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Newell et al. [39]2017.08 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Newell et al.Method category=Detection based, citation=[27]2017.10 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Chu et al.Method category=Detection based, citation=[13]2017.10 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Newell et al. ECCV'16stacks=8 (Stacked Hourglass)2018.08 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Chu et al. CVPR'172018.08 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Chu et al.2019.01 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Newell et al.2019.01 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Newell et al. (Stacked Hourglass)2019.02 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | — | — | — | — | — | — | — | — | — | |
| Ning et al.2019.02 | 96.3 | 98.1 | 92.2 | 87.8 | 90.6 | 87.6 | 82.7 | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | |
| Chu et al.2019.02 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | — | — | — | — | — | — | — | — | — | |
| Newell et al.# Param=26M, Deployment Cost=55G2018.11 | 96.3 | 98.2 | 91.2 | 87.1 | 90.1 | 87.4 | 83.6 | — | — | — | — | 90.9 | 62.9 | — | — | — | — | — | — | — | — | |
| Ning et al.# Param=74M, Deployment Cost=124G2018.11 | 96.3 | 98.1 | 92.2 | 87.8 | 90.6 | 87.6 | 82.7 | — | — | — | — | 91.2 | 63.6 | — | — | — | — | — | — | — | — | |
| Chu et al.# Param=58M, Deployment Cost=128G2018.11 | 96.3 | 98.5 | 91.9 | 88.1 | 90.6 | 88 | 85 | — | — | — | — | 91.5 | 63.8 | — | — | — | — | — | — | — | — |