Did the released model trained in cityscapes use the coarse data for training?

[xurannlpr]

Did the released pspnet101 cityscapes model use the coarse annotation data during training?

[Jinming-Su]

I think yes, because from the paper published at CVPR, the testing set is 78.4 mIoU without coarse data, and 80.2 mIoU with coarse data on Cityscapes testing set.