Keypoint Detection on COCO (test-dev)
76.2APDarkPose
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DarkPoseInput size=384x288, #Params=63.6M, FLOPs=32.9G, FPS=252020.12 | 76.2 | 92.5 | 83.6 | 72.5 | 82.4 | — | — | — | — | — | |
| HRNet-W48Input size=384x288, #Params=63.6M, FLOPs=32.9G, FPS=252020.12 | 75.5 | 92.5 | 83.3 | 71.9 | 81.5 | — | — | — | — | — | |
| TransPose-H-A6Input size=256x192, #Params=17.5M, FLOPs=21.8G, FPS=382020.12 | 75 | 92.2 | 82.3 | 71.3 | 81.1 | — | — | — | — | — | |
| HRNet-W32Input size=384x288, #Params=28.5M, FLOPs=16.0G, FPS=262020.12 | 74.9 | 92.5 | 82.8 | 71.3 | 80.9 | — | — | — | — | — | |
| TransPose-H-A4Input size=256x192, #Params=17.3M, FLOPs=17.5G, FPS=412020.12 | 74.7 | 91.9 | 82.2 | 71.4 | 80.7 | — | — | — | — | — | |
| HRNet-W48Input size=256x192, #Params=63.6M, FLOPs=14.6G, FPS=272020.12 | 74.2 | 92.4 | 82.4 | 70.9 | 79.7 | — | — | — | — | — | |
| SimpleBaselineInput size=384x288, #Params=68.6M, FLOPs=35.6G2020.12 | 73.7 | 91.9 | 81.1 | 70.3 | 80 | — | — | — | — | — | |
| TransPose-H-SInput size=256x192, #Params=8.0M, FLOPs=10.2G, FPS=452020.12 | 73.4 | 91.6 | 81.1 | 70.1 | 79.3 | — | — | — | — | — | |
| MegviiApproach=Top-down2018.07 | 73 | 91.7 | 80.9 | 69.5 | 78.1 | 79 | 95.1 | 85.9 | 74.8 | 84.6 | |
| CFNApproach=Top-down, FPS=32018.07 | 72.6 | 86.7 | 69.7 | 78.3 | 64.1 | — | — | — | — | — | |
| RMPEInput size=320x256, #Params=28.1M, FLOPs=26.7G2020.12 | 72.3 | 89.2 | 79.1 | 68 | 78.6 | — | — | — | — | — | |
| CPNInput size=384x288, #Params=58.8M, FLOPs=29.2G2020.12 | 72.1 | 91.4 | 80 | 68.7 | 77.2 | — | — | — | — | — | |
| MultiPoseNetApproach=Bottom-up, FPS=232018.07 | 69.6 | 86.3 | 76.6 | 65 | 76.3 | 73.5 | 88.1 | 79.5 | 68.6 | 80.3 | |
| Mask R-CNNApproach=Top-down, FPS=52018.07 | 69.2 | 90.4 | 76 | 64.9 | 76.3 | 75.2 | 93.7 | 81.1 | 70.3 | 81.8 | |
| SJTUApproach=Top-down, FPS=0.42018.07 | 68.8 | 87.5 | 75.9 | 64.6 | 75.1 | 73.6 | 91 | 79.8 | 68.9 | 80.2 | |
| PersonLabBackbone=ResNet152, Inference Scale=multi-scale2018.03 | 68.7 | 89 | 75.4 | 64.1 | 75.5 | 75.4 | 92.7 | 81.2 | 69.7 | 83 | |
| G-RMITraining Data=COCO+int2017.01 | 68.5 | 87.1 | 75.5 | 65.8 | 73.3 | 73.3 | 90.1 | 79.5 | 68.1 | 80.4 | |
| PersonLabBackbone=ResNet101, Inference Scale=multi-scale2018.03 | 67.8 | 88.6 | 74.4 | 63 | 74.8 | 74.5 | 92.2 | 80.4 | 68.6 | 82.5 | |
| IntegralInput size=256x256, #Params=45.0M, FLOPs=11.0G2020.12 | 67.8 | 88.2 | 74.8 | 63.9 | 74 | — | — | — | — | — | |
| GRMI-2017Approach=Top-down2018.07 | 66.9 | 86.4 | 73.6 | 64 | 72 | 71.6 | 89.2 | 77.6 | 66.1 | 79.1 | |
| PersonLabBackbone=ResNet152, Inference Scale=single-scale2018.03 | 66.5 | 88 | 72.6 | 62.4 | 72.3 | 71 | 90.3 | 76.6 | 66.1 | 77.7 | |
| PersonLab2019.04 | 66.5 | 85.5 | 71.3 | 62.3 | 70 | — | — | — | — | — | |
| Newell et al.Approach=Bottom-up, FPS=62018.07 | 65.5 | 86.8 | 72.3 | 60.6 | 72.6 | 70.2 | 89.5 | 76 | 64.6 | 78.1 | |
| Associative EmbeddingMethod Category=Bottom-up, Inference Scale=multi-scale, Refinement=Yes2018.03 | 65.5 | 87.9 | 77.7 | 69 | 75.2 | 75.8 | 91.2 | 81.9 | 71.4 | 82 | |
| PersonLabBackbone=ResNet101, Inference Scale=single-scale2018.03 | 65.5 | 87.1 | 71.4 | 61.3 | 71.5 | 70.1 | 89.7 | 75.7 | 65 | 77.1 | |
| G-RMITraining Data=COCO-only2017.01 | 64.9 | 85.5 | 71.3 | 62.3 | 70 | 69.7 | 88.7 | 75.5 | 64.4 | 77.1 | |
| G-RMI COCO-onlyMethod Category=Top-down2018.03 | 64.9 | 85.5 | 71.3 | 62.3 | 70 | 69.7 | 88.7 | 75.5 | 64.4 | 77.1 | |
| G-RMIInput size=353x257, #Params=42.6M, FLOPs=57G2020.12 | 64.9 | 85.5 | 71.3 | 62.3 | 70 | — | — | — | — | — | |
| Mask-RCNN2017.01 | 63.1 | 87.3 | 68.7 | 57.8 | 71.4 | — | — | — | — | — | |
| Mask R-CNN, keypoint & maskBackbone=ResNet-50-FPN, Multi-task=keypoint & mask2017.03 | 63.1 | 87.3 | 68.7 | 57.8 | 71.4 | — | — | — | — | — | |
| Mask-RCNNMethod Category=Top-down2018.03 | 63.1 | 87.3 | 68.7 | 57.8 | 71.4 | 69.7 | 91.6 | 74.9 | 63.7 | 77.8 | |
| Mask-RCNN2019.04 | 63.1 | 87.3 | 68.7 | 57.8 | 71.4 | — | — | — | — | — | |
| Associative EmbeddingMethod Category=Bottom-up, Inference Scale=multi-scale2018.03 | 63 | 85.7 | 68.9 | 58 | 70.4 | — | — | — | — | — | |
| CenterNet (HG-jd)Backbone=Hourglass, Strategy=Matching regression to the closest joint detection2019.04 | 63 | 86.8 | 69.6 | 58.9 | 70.4 | — | — | — | — | — | |
| Pose-AE2019.04 | 62.8 | 84.6 | 69.2 | 57.5 | 70.6 | — | — | — | — | — | |
| Mask R-CNN, keypoint-onlyBackbone=ResNet-50-FPN, Multi-task=keypoint-only2017.03 | 62.7 | 87 | 68.4 | 57.4 | 71.1 | — | — | — | — | — | |
| G-RMITraining data=COCO plus MPII, Backbone (bbox)=Inception-ResNet-v2, Backbone (keypoints)=ResNet-1012017.03 | 62.4 | 84 | 68.5 | 59.1 | 68.1 | — | — | — | — | — | |
| CMU-Pose2017.01 | 61.8 | 84.9 | 67.5 | 57.1 | 68.2 | 66.5 | 87.2 | 71.8 | 60.6 | 74.6 | |
| CMU-PoseApproach=Bottom-up, FPS=102018.07 | 61.8 | 84.9 | 67.5 | 57.1 | 68.2 | 66.5 | 87.2 | 71.8 | 60.6 | 74.6 | |
| CMU-Pose+++Multi-scale testing=true, Post-processing=CPM, Filtering=object detector2017.03 | 61.8 | 84.9 | 67.5 | 57.1 | 68.2 | — | — | — | — | — | |
| CMU-PoseMethod Category=Bottom-up, Refinement=Yes2018.03 | 61.8 | 84.9 | 67.5 | 57.1 | 68.2 | 66.5 | 87.2 | 71.8 | 60.6 | 74.6 | |
| CMU-Pose2019.04 | 61.8 | 84.9 | 67.5 | 58 | 70.4 | — | — | — | — | — | |
| G-RMI-2016Approach=Top-down2018.07 | 60.5 | 82.2 | 66.2 | 57.6 | 66.6 | 66.2 | 86.6 | 71.4 | 61.9 | 72.2 | |
| CenterNet (DLA-jd)Backbone=DLA, Strategy=Matching regression to the closest joint detection2019.04 | 57.9 | 84.7 | 63.1 | 52.5 | 67.4 | — | — | — | — | — | |
| CenterNet (HG-reg)Backbone=Hourglass, Strategy=Direct center-out offset regression2019.04 | 55 | 83.5 | 59.7 | 49.4 | 64 | — | — | — | — | — | |
| CenterNet (DLA-reg)Backbone=DLA, Strategy=Direct center-out offset regression2019.04 | 51.7 | 81.4 | 55.2 | 44.6 | 63 | — | — | — | — | — |