jakeret
Optimization of last convolutional layer
Hello!
I've just noticed that "weight" and "bias" from the last convolutional layer are not included in "variables", aren't they supposed to be optimized as well? Is there any reason for not including them?
Best regards, Amelia.
You're right, this incorrect. Luckily, this has no effect on the network. The 'variables' are only used to compute the normalized gradients in order to visuize the overall training process. If you have time a would greatly appreciate a PR that fixes the issue ;-)
On Jun 13, 2017 19:29, "amejs" [email protected] wrote:
Hello!
I've just noticed that "weight" and "bias" from the last convolutional layer are not included in "variables", aren't they supposed to be optimized as well? Is there any reason for not including them?
Best regards, Amelia.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jakeret/tf_unet/issues/62, or mute the thread https://github.com/notifications/unsubscribe-auth/ALSFv1bVqvXmSI5ITvBdTdMzm5Pi7gpjks5sDsbjgaJpZM4N40Hu .
