https://github.com/tensorflow/tensorflow/pull/18098 patch for tf1.4

gradient_penalty.patch

commit 1ebd8b170d31d64ad2523a1db81c5619fed24fc1
Author: Lukas Geiger <[email protected]>
Date:   Sat Mar 31 01:40:36 2018 +0200

    one_sided penalty

diff --git a/tensorflow/contrib/gan/python/losses/python/losses_impl.py b/tensorflow/contrib/gan/python/losses/python/losses_impl.py
index 2a40dbade6..77a86043d8 100644
--- a/tensorflow/contrib/gan/python/losses/python/losses_impl.py
+++ b/tensorflow/contrib/gan/python/losses/python/losses_impl.py
@@ -297,6 +297,7 @@ def wasserstein_gradient_penalty(
     discriminator_fn,
     discriminator_scope,
     epsilon=1e-10,
+    one_sided=False,
     weights=1.0,
     scope=None,
     loss_collection=ops.GraphKeys.LOSSES,
@@ -364,10 +365,13 @@ def wasserstein_gradient_penalty(
   # For numerical stability, add epsilon to the sum before taking the square
   # root. Note tf.norm does not add epsilon.
   slopes = math_ops.sqrt(gradient_squares + epsilon)
-  penalties = math_ops.square(slopes - 1.0)
+  penalties = slopes - 1.0
+  if one_sided:
+    penalties = math_ops.maximum(0., penalties)
+  penalties_squared = math_ops.square(penalties)
   penalty = losses.compute_weighted_loss(
-      penalties, weights, scope=scope, loss_collection=loss_collection,
-      reduction=reduction)
+      penalties_squared, weights, scope=scope,
+      loss_collection=loss_collection, reduction=reduction)
 
   if add_summaries:
     summary.scalar('gradient_penalty_loss', penalty)
diff --git a/tensorflow/contrib/gan/python/train.py b/tensorflow/contrib/gan/python/train.py
index 06dd281489..f72e10f060 100644
--- a/tensorflow/contrib/gan/python/train.py
+++ b/tensorflow/contrib/gan/python/train.py
@@ -334,6 +334,7 @@ def gan_loss(
     # Auxiliary losses.
     gradient_penalty_weight=None,
     gradient_penalty_epsilon=1e-10,
+    gradient_penalty_one_sided=False,
     mutual_information_penalty_weight=None,
     aux_cond_generator_weight=None,
     aux_cond_discriminator_weight=None,
@@ -406,7 +407,10 @@ def gan_loss(
   # Add optional extra losses.
   if _use_aux_loss(gradient_penalty_weight):
     gp_loss = tfgan_losses.wasserstein_gradient_penalty(
-        model, epsilon=gradient_penalty_epsilon, add_summaries=add_summaries)
+        model,
+        epsilon=gradient_penalty_epsilon,
+        one_sided=gradient_penalty_one_sided,
+        add_summaries=add_summaries)
     dis_loss += gradient_penalty_weight * gp_loss
   if _use_aux_loss(mutual_information_penalty_weight):
     info_loss = tfgan_losses.mutual_information_penalty(