Fixed a problem with tensorflow static graph mode. The loss node is connected to a layer whose name starts with "semantic_" as this name indicated the last layer of the deepcell models. I renamed the last layer of the unet "semantic_head" instead of "outputs". In addition I change the last layer activation from softmax to sigmoid, since we do binary classification.
What is the purpose of this PR?
Fixed a problem with tensorflow static graph mode. The loss node is connected to a layer whose name starts with "semantic_" as this name indicated the last layer of the deepcell models. I renamed the last layer of the unet "semantic_head" instead of "outputs". In addition I change the last layer activation from softmax to sigmoid, since we do binary classification.
How did you implement your changes
Changed two lines of code.
Remaining issues None