facebookresearch/Mask2Former

DataSet

zhaofh1115 opened this issue · 1 comments

Hello, how is your data set labeled? I used the 1 classification task data set, labeled as: background 0, target 1. After the processing in your way, it becomes background 255, target 0. Then it is used for training, but it seems that the training result is faulty

If you are referring to semantic segmentation, a label of 255 is usually ignored in most dataset (see ignored_label).