search

speedinghzl / CCNet

CCNet: Criss-Cross Attention for Semantic Segmentation (TPAMI 2020 & ICCV 2019).
MIT License
1.42k stars 277 forks source link

About road segmentation #20

Closed Amore-HDU closed 5 years ago

Amore-HDU commented 5 years ago

Hello， Now I try to use CCNet to segment Road. In my mask images, road is set 255 while others are set 0. NUM_CLASSES=2, self.id_to_trainid={0:0, 255:255}. It quickly convergence.But when I run test.py,the output image is all black. I want to knows what's wrong? Thx @speedinghzl

speedinghzl commented 5 years ago

255 is used as ignore label in this repo. You can use this mapping {0:0, 255:1}