Multi-Instance Pose Networks: Rethinking Top-Down Pose Estimation

Benchmark Model Rank Results
2d-human-pose-estimation-on-ochumanMIPNet (HRNet-W48)#3Test AP: 42.5Validation AP: 42.0
2d-human-pose-estimation-on-ochumanHRNet-W48#4Test AP: 37.2Validation AP: 37.8
keypoint-detection-on-cocoMIPNet(384x288)#7Test AP: 75.7Validation AP: 76.3
keypoint-detection-on-ochumanMIPNet (HRNet-W48)#2Test AP: 42.5Validation AP: 42.0
keypoint-detection-on-ochumanHRNet-W48#3Test AP: 37.2Validation AP: 37.8
multi-person-pose-estimation-on-crowdposeMIPNet (HRNet-W48)#11mAP @0.5:0.95: 70.0AP Easy: 78.1AP Medium: 71.1AP Hard: 59.4
multi-person-pose-estimation-on-ochumanMIPNet (gt-bb)#1AP50: 89.7AP75: 80.1Validation AP: 74.1
pose-estimation-on-coco-test-devMIPNet#19AP: 75.7AP50: 92.4AP75: 83.3APL: 81.2APM: 71.4AR: 80.5
pose-estimation-on-crowdposeMIPNet (HRNet-W48)#8AP: 70.0APM: 71.1AP Hard: 59.4
pose-estimation-on-ochumanMIPNet (HRNet-W48)#10Test AP: 42.5Validation AP: 42.0
pose-estimation-on-ochumanHRNet-W48#12Test AP: 37.2Validation AP: 37.8