weigq/3d_pose_baseline_pytorch

A bug in reading the test data

Opened this issue · 1 comments

When the required action is sittingdown, the current file will read both the sitting and sittingdown. This will affect the accuracy of the sequence of sittingdown
So I suggest to add some lines to discriminate these two kinds of actions after the line 61 in the human36m.py.

if self.actions=='SittingDown' and act.endwith('g'):
continue

weigq commented

Good, i will check that.