MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video

Benchmark Model Rank Results
3d-human-pose-estimation-on-human36mMixSTE (HRNet, T=243)#8Average MPJPE (mm): 39.8Using 2D ground-truth joints: No
3d-human-pose-estimation-on-human36mMixSTE (CPN, T=243)#10Average MPJPE (mm): 40.9Using 2D ground-truth joints: No
3d-human-pose-estimation-on-human36mMixSTE (CPN, T=81)#13Average MPJPE (mm): 42.4Using 2D ground-truth joints: No
3d-human-pose-estimation-on-humaneva-iMixSTE (T=43, FT)#3Mean Reconstruction Error (mm): 16.1
3d-human-pose-estimation-on-mpi-inf-3dhpMixSTE (T=27)#18MPJPE: 54.9AUC: 66.5PCK: 94.4
3d-human-pose-estimation-on-mpi-inf-3dhpMixSTE (T=1)#20MPJPE: 57.9AUC: 63.8PCK: 94.2
classification-on-full-body-parkinsonsMixste#7F1-score (weighted): 0.41
monocular-3d-human-pose-estimation-on-human3MixSTE (HRNet, T=243)#7Average MPJPE (mm): 39.8Use Video Sequence: YesFrames Needed: 243