(Excuse me if a similar issue is posted earlier - I couldn't find one).
Is there a chance that the ground truth mask labels reported on face_parsing/README.md are wrong?
If you run g_mask.py to "fuse" the different mask images together, the resulting mask labels are different.
e.g. In the label list, the reported label for nose is 2 while in the resulting "fused" mask the nose part has the value 10.
Hi,
The label list defines the fused order like making a pizza. Therefore, it is possible that the before-ordered labels are overwritten by the latter-ordered labels.
Hi and thanks for this great dataset,
(Excuse me if a similar issue is posted earlier - I couldn't find one). Is there a chance that the ground truth mask labels reported on face_parsing/README.md are wrong? If you run
g_mask.py
to "fuse" the different mask images together, the resulting mask labels are different.e.g. In the label list, the reported label for nose is
2
while in the resulting "fused" mask the nose part has the value10
.Thanks in advance. Antonis