Change loss function to use tensorflow defaults

[meetps]

Is there any specific reason why you chose to write a custom loss function instead of directly using this?

# mask_.shape == (batch_size, h, w, n_classes)
# y_mask_.shape == (batch_size*h*w, n_classes)
mask_ =  tf.reshape(mask_, (-1, n_classes))
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=mask_, labels=y_mask))

I am wondering if there is any added advantage by using your loss function.

[MarvinTeichmann]

Yes, I ran into numerical instabilities with the tensorflow loss function. For details also see this issue.