arifkhan21 / mpii-dataset-in-csv Goto Github PK

I was looking at the .mat file myself and was very confused as to why there were inconsistencies to the data. For example every annotation may not have the same subheaders/have any joint coordinates. After looking at your work, I noticed you did some preprocessing to get the neat csv. Have you ever figured why there were such "blanks" in the original dataset/why it was structured this way?

Case in point:
if 'annopoints' not in obj_rect._fieldnames: continue
In this line above it seems like you were skipping items that didn't have joint annotations.

Thank you and I look forward to your reply.