SwapNet icon indicating copy to clipboard operation
SwapNet copied to clipboard

Labels are off by 1, hats are missing

Open gauravkuppa opened this issue 4 years ago • 1 comments

It seems that most of the labels (except hair) are off by 1. They deviate from the cloth labels in decode_labels.py. For example, face is said to be 13, according to the labels, but in the cloth segmentation numpy array, the face is represented by 12.

Additionally, hats are said to be 1 in the labels, and the issue with the labels being off by 1 causes the hats to be considered the background.

02_1_front.jpg image 02_1_front.npz image

gauravkuppa avatar Apr 12 '20 02:04 gauravkuppa

Update: sunglasses are not labeled either

gauravkuppa avatar Apr 12 '20 03:04 gauravkuppa