search

hellochick / ICNet-tensorflow

TensorFlow-based implementation of "ICNet for Real-Time Semantic Segmentation on High-Resolution Images".
406 stars 153 forks source link

question about the pretrained model #21

Closed glynpu closed 6 years ago

glynpu commented 6 years ago

The dataset has 19 classes, plus the background, the output of last layer should be 20 classes. But the pre-trained model produces a 19 channels output. Do the background just be ignored? But in the train code, the ignored-label is set to 0.
I am a little confused, please help!

hellochick commented 6 years ago

Hey, sorry for confusing you. The ignore-label of cityscapes is not 0 (is 255), I forgot to change it before I upload the training code. So when training, we will focus on the 0~18 classes and ignore the background (255), which means we don't want to learn how to recognize the background. Does that make sense to you?