Join GitHub today
GitHub is home to over 36 million developers working together to host and review code, manage projects, and build software together.Sign up
DL4J: UNet pretrained model configuration issue #6955
Has sigmoid going into softmax... this is almost certainly wrong, unless there's some unusual reason for doing this in UNet...
I think the following changes should be made to the model: Remove the last conv layer ("conv10") which outputs 1 feature map (this layer does not exist in the original paper) and change the activation function in the new last conv layer ("conv9-3") to identity with two output feature maps (which is the same as in the original paper) (the number of output feature maps will correspond to the number of segmentation classes).