Human Keypoint Detection on COCO
78.9APCCM
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CCMCategory=Top down, Backbone=ResNet-152, HRNet-w48, Decoder=CCM, Extra Data=true, Post-processing=sub-pixel refinement2020.02 | 78.9 | — | — | — | — | — | — | |
| SOTA (TD)Paradigm=Top-Down, Flipping test=true2023.10 | 78.6 | — | — | — | — | — | — | |
| Li et al. (2019)Category=Top down, Backbone=4 x ResNet-50, Decoder=GlobalNet, Extra Data=false, Post-processing=flip/GF/sub-pixel shift2020.02 | 78.1 | — | — | — | — | — | — | |
| HRFormer2022.12 | 77.2 | — | — | — | — | — | — | |
| Sun et al. (2019)Category=Top down, Backbone=HRNet-w48, Decoder=1x1 conv, Extra Data=false, Post-processing=flip/sub-pixel shift2020.02 | 77 | — | — | — | — | — | — | |
| X-Pose-T (E2E)Backbone=Swin-L, Prompt Type=Textual2023.10 | 76.8 | — | — | — | — | — | — | |
| X-Pose-V (E2E)Backbone=Swin-L, Prompt Type=Visual2023.10 | 76.6 | — | — | — | — | — | — | |
| Xiao et al. (2018)Category=Top down, Backbone=ResNet-152, Decoder=deconv, Extra Data=false, Post-processing=flip/sub-pixel shift2020.02 | 76.5 | — | — | — | — | — | — | |
| HRNet2022.12 | 76.3 | — | — | — | — | — | — | |
| X-Pose-T (E2E)Backbone=Swin-T, Prompt Type=Textual2023.10 | 74.4 | — | — | — | — | — | — | |
| X-Pose-V (E2E)Backbone=Swin-T, Prompt Type=Visual2023.10 | 74.3 | — | — | — | — | — | — | |
| Chen et al. (2018b)Category=Top down, Backbone=ResNet-Inception, Decoder=GlobalNet, Extra Data=false, Post-processing=flip/GF2020.02 | 72.1 | — | — | — | — | — | — | |
| PainterBackbone=ViT-L2022.12 | 72.1 | — | — | — | — | — | — | |
| Papandreou et al. (2018)Category=Top down, Backbone=ResNet-101, Decoder=1x1 conv, Extra Data=false, Post-processing=offset regression2020.02 | 68.5 | — | — | — | — | — | — | |
| Pix2Seq v2Training Setup=single task, Resolution=640x6402022.06 | 68 | — | — | — | — | — | — | |
| Pix2Seq v2Training Setup=single task, Resolution=1024x10242022.06 | 67.4 | — | — | — | — | — | — | |
| Mask R-CNN (non-local)2022.06 | 66.5 | — | — | — | — | — | — | |
| DeCon-ML-LPret. Dataset=COCO, Backbone=ResNet-50, Pret. Dec.=FPN2025.03 | 65.88 | — | — | — | — | — | — | |
| DeCon-SLPret. Dataset=COCO, Backbone=ResNet-50, Pret. Dec.=FCN2025.03 | 65.77 | — | — | — | — | — | — | |
| SlotCon (2022)Pret. Dataset=COCO, Backbone=ResNet-50, Pret. Dec.=-2025.03 | 65.66 | — | — | — | — | — | — | |
| DeCon-ML-SPret. Dataset=COCO, Backbone=ResNet-50, Pret. Dec.=FPN2025.03 | 65.66 | — | — | — | — | — | — | |
| Newell et al. (2017)Category=Bottom-up, Backbone=Hourglass, Decoder=-, Extra Data=false, Post-processing=multiscale average2020.02 | 65.5 | — | — | — | — | — | — | |
| Pix2Seq v2Training Setup=multi-tasks, Resolution=640x6402022.06 | 65 | — | — | — | — | — | — | |
| Pix2Seq v2Training Setup=multi-tasks, Resolution=1024x10242022.06 | 64.8 | — | — | — | — | — | — | |
| Pix2Seq v2Backbone=ViT-B2022.12 | 64.8 | — | — | — | — | — | — | |
| Random init.Pret. Dataset=-, Backbone=ResNet-50, Pret. Dec.=-2025.03 | 63.92 | — | — | — | — | — | — | |
| Fang et al. (2017)Category=Top down, Backbone=STN+Hourglass, Decoder=-, Extra Data=false, Post-processing=parametric NMS2020.02 | 63.3 | — | — | — | — | — | — | |
| He et al. (2017)Category=Top down, Backbone=ResNet-50-FPN, Decoder=conv+deconv, Extra Data=false, Post-processing=offset regression2020.02 | 63.1 | — | — | — | — | — | — | |
| Mask R-CNN2022.06 | 63.1 | — | — | — | — | — | — | |
| Cao et al. (2017)Category=Bottom-up, Backbone=VGG-19, Decoder=multi-stage CNN, Extra Data=false, Post-processing=-2020.02 | 61.8 | — | — | — | — | — | — | |
| CLIFF#Params=78.89, FPS=22.422024.06 | — | 33.2 | — | — | 49.3 | 61.4 | 72.5 | |
| DaNet#Params=102.25, FPS=11.952024.06 | — | 19.8 | — | — | 36.2 | 52.6 | 65.2 | |
| FastMetro#Params=153.74, FPS=16.042024.06 | — | 13.6 | — | — | 28 | 39.3 | 53.2 | |
| Graphormer#Params=226.21, FPS=14.422024.06 | — | 12 | — | — | 25.7 | 35.9 | 49.5 | |
| HybrIK#Params=75.69, FPS=7.572024.06 | — | 10.6 | — | — | 24.9 | 31.7 | 47.5 | |
| JCLFramework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=2002021.12 | — | 66.2 | 87.2 | 72.3 | — | — | — | |
| MeshPoseBackbone=HRNet32, #Params=46.29, FPS=28.662024.06 | — | 47.3 | — | — | 61.3 | 71.2 | 79.6 | |
| MeshPoseSBackbone=ResNet50, #Params=45.37, FPS=124.282024.06 | — | 43.8 | — | — | 58.1 | 67 | 76.5 | |
| MeshPoseXSBackbone=MBNet140, #Params=21.25, FPS=124.212024.06 | — | 40.6 | — | — | 55.2 | 63.6 | 73.9 | |
| Metro#Params=243.07, FPS=15.472024.06 | — | 9.5 | — | — | 22.4 | 30.4 | 45.1 | |
| MoCo-v1Framework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=2002021.12 | — | 65.6 | 87.1 | 71.3 | — | — | — | |
| MoCo-v2Framework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=2002021.12 | — | 66 | 87.2 | 71.4 | — | — | — | |
| NIKI#Params=92.17, FPS=6.182024.06 | — | 21.3 | — | — | 37.2 | 48.5 | 61.5 | |
| PARE#Params=32.86, FPS=20.482024.06 | — | 33.2 | — | — | 48.1 | 56.9 | 67.2 | |
| PointHMR#Params=59.09, FPS=15.382024.06 | — | 17.7 | — | — | 32.7 | 44.4 | 57.3 | |
| PyMaf#Params=45.18, FPS=35.572024.06 | — | 23 | — | — | 40.4 | 58 | 70 | |
| randomFramework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=02021.12 | — | 63.5 | 85.3 | 69.3 | — | — | — | |
| supervisedFramework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=N/A2021.12 | — | 65.4 | 87 | 71 | — | — | — | |
| SwAVFramework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=2002021.12 | — | 65.6 | 86.9 | 71.6 | — | — | — | |
| SwAV+UOTAFramework=Keypoint R-CNN, Backbone=ResNet-50, FPN=true, Finetuning schedule=1x, Pre-training epochs=2002021.12 | — | 66.3 | 87.4 | 72.3 | — | — | — |