bearpaw/pytorch-pose

detections_our_format

nitba opened this issue · 2 comments

nitba commented

Hi @bearpaw

which script 'evaluation/data/detections_our_format.mat' produce this result? and jnt_missing attribute is the same as the attribute `` in data loader during training, I could not find that, I mean the only joints which has zero coordinates in their annotation are considered as missed ones ? not occluded ones?

nitba commented

Why do you have SC_BIAS in addition to '0.5 threshold'? SC_BIAS is considered for previous State of arts?

bearpaw commented
  1. detections_our_format.mat is from http://www.cims.nyu.edu/~tompson/data/mpii_valid_pred.zip (see Readme)
  2. Eval code is from MPII's official evaluation toolkit: http://human-pose.mpi-inf.mpg.de/results/mpii_human_pose/evalMPII.zip