Keypoint Detection on MS-COCO 2017 (val)
75.8APHRNet
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| HRNetApproach=Heatmap based, Backbone=HRNet-W32, Input size=384 x 288, #Params=28.5M, GFLOPS=162021.04 | 75.8 | 90.6 | 82.7 | 71.9 | 82.8 | 81 | — | — | |
| SimpleBaselineApproach=Heatmap based, Backbone=ResNet-101, Input size=384 x 288, #Params=53.0M, GFLOPS=26.72021.04 | 73.6 | 89.6 | 80.3 | 69.9 | 81.1 | 79.1 | — | — | |
| PRTRApproach=Regression based, Backbone=HRNet-W32, Input size=512 x 384, #Params=57.2M, GFLOPS=37.82021.04 | 73.3 | 89.2 | 79.9 | 69 | 80.9 | 80.2 | — | — | |
| PRTRApproach=Regression based, Backbone=HRNet-W32, Input size=384 x 288, #Params=57.2M, GFLOPS=21.62021.04 | 73.1 | 89.4 | 79.8 | 68.8 | 80.4 | 79.8 | — | — | |
| SimpleBaselineApproach=Heatmap based, Backbone=ResNet-50, Input size=384 x 288, #Params=34.0M, GFLOPS=18.62021.04 | 72.2 | 89.3 | 78.9 | 68.1 | 79.7 | 77.6 | — | — | |
| PRTRApproach=Regression based, Backbone=ResNet-101, Input size=512 x 384, #Params=60.4M, GFLOPS=33.42021.04 | 72 | 89.3 | 79.4 | 67.3 | 79.7 | 79.2 | — | — | |
| PRTRApproach=Regression based, Backbone=ResNet-50, Input size=512 x 384, #Params=41.5M, GFLOPS=18.82021.04 | 71 | 89.3 | 78 | 66.4 | 78.8 | 78 | — | — | |
| PRTRApproach=Regression based, Backbone=ResNet-101, Input size=384 x 288, #Params=60.4M, GFLOPS=19.12021.04 | 70.1 | 88.8 | 77.6 | 65.7 | 77.4 | 77.5 | — | — | |
| PointSetNetApproach=Regression based, Backbone=HRNet-W48, multi-scale test=true2021.04 | 69.8 | 88.8 | 76.3 | — | — | — | — | — | |
| CPNApproach=Heatmap based, Backbone=ResNet-50, Input size=256 x 192, #Params=27.0M, GFLOPS=6.22021.04 | 68.6 | — | — | — | — | — | — | — | |
| PRTRApproach=Regression based, Backbone=ResNet-50, Input size=384 x 288, #Params=41.5M, GFLOPS=112021.04 | 68.2 | 88.2 | 75.2 | 63.2 | 76.2 | 76 | — | — | |
| NASNetParams (M)=10.66, FLOPS (M)=569.11, Flip during validation=true, Framework=SimpleBaseline2021.08 | 67.9 | — | — | — | — | — | — | — | |
| MoCopre-train=MoCo IG-1B2019.11 | 66.9 | 87.8 | 73 | — | — | — | — | — | |
| 8-stage HourglassApproach=Heatmap based, Backbone=Hourglass-8 stacked, Input size=256 x 192, #Params=25.1M, GFLOPS=14.32021.04 | 66.9 | — | — | — | — | — | — | — | |
| DARTSParams (M)=9.20, FLOPS (M)=531.77, Flip during validation=true, Framework=SimpleBaseline2021.08 | 66.9 | — | — | — | — | — | — | — | |
| MoCopre-train=MoCo IN-1M2019.11 | 66.8 | 87.4 | 72.5 | — | — | — | — | — | |
| SetSimArchitecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.7 | 87.8 | 72.4 | — | — | — | — | — | |
| EEEA-Net-C2Params (M)=7.47, FLOPS (M)=297.49, Flip during validation=true, Framework=SimpleBaseline2021.08 | 66.7 | — | — | — | — | — | — | — | |
| DenseCLArchitecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.6 | 87.4 | 72.6 | — | — | — | — | — | |
| SCRLpretrain=SCRL, Backbone=ResNet-50, Architecture=Mask R-CNN w/ FPN2021.03 | 66.5 | 87.8 | 72.3 | — | — | — | — | — | |
| PixProArchitecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.5 | 87.6 | 72.3 | — | — | — | — | — | |
| MoCo-v2Architecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.3 | 87.1 | 72.2 | — | — | — | — | — | |
| ReSim-C4Architecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.3 | 87.2 | 72.4 | — | — | — | — | — | |
| PRTRApproach=Regression based, Backbone=HRNet-W48, end-to-end variant=true2021.04 | 66.2 | 85.9 | 72.1 | 61.3 | 74.4 | 72.2 | — | — | |
| MoCo-v1Architecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.1 | 86.7 | 72.4 | — | — | — | — | — | |
| VADERArchitecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x, Pre-training Epochs=2002021.07 | 66.1 | 87.3 | 72.1 | — | — | — | — | — | |
| random initializationpre-train=random init.2019.11 | 65.9 | 86.5 | 71.7 | — | — | — | — | — | |
| ImageNet supervisedpre-train=super. IN-1M2019.11 | 65.8 | 86.9 | 71.9 | — | — | — | — | — | |
| BYOLpretrain=BYOL, Backbone=ResNet-50, Architecture=Mask R-CNN w/ FPN2021.03 | 65.8 | 87 | 72 | — | — | — | — | — | |
| PointSetNetApproach=Regression based, Backbone=ResNeXt-101 DCN, multi-scale test=true2021.04 | 65.7 | 85.4 | 71.8 | — | — | — | — | — | |
| supervised-INpretrain=supervised-IN, Backbone=ResNet-50, Architecture=Mask R-CNN w/ FPN2021.03 | 65.7 | 87.1 | 71.7 | — | — | — | — | — | |
| IN-1K sup.Architecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x2021.07 | 65.3 | 87 | 71.3 | — | — | — | — | — | |
| MobileNet-V3Params (M)=9.01, FLOPS (M)=223.16, Flip during validation=true, Framework=SimpleBaseline2021.08 | 65.3 | — | — | — | — | — | — | — | |
| MobileNet-V2Params (M)=9.57, FLOPS (M)=306.80, Flip during validation=true, Framework=SimpleBaseline2021.08 | 64.9 | — | — | — | — | — | — | — | |
| PRTRApproach=Regression based, Backbone=ResNet-101, end-to-end variant=true2021.04 | 64.8 | 85.1 | 70.2 | 60.4 | 73.8 | 73.9 | — | — | |
| HG-RCNNBackbone=ResNeXt-1012019.09 | 63.48 | 86.2 | 69.05 | 58.4 | 72.04 | — | — | — | |
| randompretrain=random, Backbone=ResNet-50, Architecture=Mask R-CNN w/ FPN2021.03 | 63.2 | 85.3 | 68.9 | — | — | — | — | — | |
| Random init.Architecture=Keypoint-RCNN, Backbone=ResNet-50, Neck=FPN, Fine-tuning Schedule=1x2021.07 | 63 | 85.1 | 68.4 | — | — | — | — | — | |
| MnasNetParams (M)=10.45, FLOPS (M)=320.17, Flip during validation=true, Framework=SimpleBaseline2021.08 | 62.5 | — | — | — | — | — | — | — | |
| ShuffleNet-V2Params (M)=7.55, FLOPS (M)=154.37, Flip during validation=true, Framework=SimpleBaseline2021.08 | 60.4 | — | — | — | — | — | — | — | |
| Multi-HMR 2.bTraining with 2D annotations=true2026.06 | 50.7 | 81 | 54.3 | — | — | 56.2 | 83.2 | 60.9 | |
| AiOSTraining with 2D annotations=false2026.06 | 44.9 | 76 | 46.4 | — | — | 54 | 83.3 | 57.7 | |
| SAT-HMRTraining with 2D annotations=true2026.06 | 42.1 | 73.8 | 43.8 | — | — | 49.5 | 78.4 | 53 | |
| Multi-HMRTraining with 2D annotations=false2026.06 | 31.2 | 60 | 29.1 | — | — | 39 | 66 | 39.8 |