List your company

Multi-Person Pose Estimation on MPII (full test)

92.2Head Joint AccuracyGPN
57.04866.17475.384.426May 10, 2016Nov 26, 2016Jun 14, 2017Dec 31, 2017Jul 19, 2018Feb 4, 2019Aug 24, 2019

Evaluation results

MethodLinks
GPNTime [s]=0.772017.0592.289.782.174.478.676.469.380.4—
Associative Embedding (Our method)Refinement=Single-person pose model [40], Scales=Multi-scale, Backbone=Four stacked hourglass modules2016.1192.189.378.969.876.271.664.777.5—
Newell and Deng2017.0592.189.378.969.876.271.664.777.5—
Newell & Deng2016.1292.189.378.969.876.271.664.777.5—
Newell and DengTime[s]=0.252019.0892.189.378.969.876.271.664.777.5—
Fieraru et al.2019.0891.889.580.469.677.371.765.578—
BU-fullSetting=BU-full2016.1291.587.874.662.572.265.356.772.90.12
RMPE++Human detector=Faster-RCNN with soft-nms [5], Pose estimator=PyraNet [45], Input size=320x2562016.1291.390.58476.480.379.972.482.1—
Part Affinity FieldsScale Search=true, s/image=0.0052016.1191.287.677.766.875.468.961.775.6—
Cao et al.venue=CVPR172016.1191.287.677.766.875.468.961.775.6—
Cao et al.Time [s]=1.242017.0591.287.677.766.875.468.961.775.6—
Cao et al.2016.1291.287.677.766.875.468.961.775.6—
Cao et al.Time[s]=0.62019.0891.287.677.766.875.468.961.775.6—
Levinkov et al.venue=CVPR172016.1189.885.271.859.671.16353.570.6—
Levinkov et al.2017.0589.885.271.859.671.16353.570.6—
Levinkov et al.2016.1289.885.271.859.671.16353.570.6—
Levinkov et al.2019.0889.885.271.859.671.16353.570.6—
SPMTime[s]=0.0582019.0889.787.480.472.476.774.968.378.5—
DeeperCutOptimization Stages=3-stage (incremental), Excluding out-of-group predictions=true2016.0589.484.570.459.368.962.754.670485
DeeperCutSetting=DeeperCut [15]2016.1289.484.570.459.368.962.754.670485
Insafutdinov et al. [45]venue=arXiv16a2016.1189.484.570.459.368.962.754.670—
Part Affinity FieldsScale Search=false, s/image=0.0052016.118984.974.964.27165.658.172.5—
TD/BU+ SPSetting=TD/BU+ SP2016.1288.88775.964.974.268.860.574.30.005
Insafutdinov et al. [27]venue=CVPR172016.1188.88775.964.974.268.860.574.3—
Insafutdinov et al.2017.0588.88775.964.974.268.860.574.3—
Insafutdinov et al.2016.1288.88775.964.974.268.860.574.3—
Insafutdinov et al. [15]2019.0888.88775.964.974.268.860.574.3—
Fang et al.venue=arXiv172016.1188.486.578.670.474.47365.876.7—
Fang et al.Time [s]=1.52017.0588.486.578.670.474.47365.876.7—
RMPE2016.1288.486.578.670.474.47365.876.7—
Fang et al.Time[s]=0.42019.0888.486.578.670.474.47365.876.7—
DeeperCutOptimization Stages=3-stage (incremental), Excluding out-of-group predictions=false2016.0579.172.259.750565144.659.4485
DeeperCutSetting=DeeperCut [14]2016.1279.172.259.750565144.659.4485
DeeperCuts/image=4852016.1178.472.560.25157.25245.459.5—
Insafutdinov et al. [28]venue=ECCV162016.1178.472.560.25157.25245.459.5—
Insafutdinov et al.Time [s]=4852017.0578.472.560.25157.25245.459.5—
DeeperCut2016.1278.472.560.25157.25245.459.5—
Insafutdinov et al. [16]Time[s]=4852019.0878.472.560.25157.25245.459.5—
Faster R-CNN + unaryPre-localization=Faster R-CNN [37], Excluding out-of-group predictions=true2016.0575.173.662.75161.152.642.259.71
DeeperCutOptimization Stages=1-stage, Excluding out-of-group predictions=false2016.0573.765.454.945.252.347.840.754.72,785
Faster R-CNN + unaryPre-localization=Faster R-CNN [37], Excluding out-of-group predictions=false2016.0564.962.953.444.150.743.135.2511
Iqbal et al.s/image=102016.1158.453.944.53542.236.731.143.1—
Iqbal&GallSetting=Iqbal&Gall [16]2016.1258.453.944.53542.236.731.143.110
Iqbal & Gallvenue=ECCV162016.1158.453.944.53542.236.731.143.1—
Iqbal and GallTime [s]=102017.0558.453.944.53542.236.731.143.1—
Iqbal & Gall2016.1258.453.944.53542.236.731.143.1—
Iqbal and GallTime[s]=102019.0858.453.944.53542.236.731.143.1—