Add the missing is_training to batch_norm and dropout.

research/attention_ocr/python/model.py

@@ -201,6 +201,8 @@ class Model(object):
     with tf.variable_scope('conv_tower_fn/INCE'):
       if reuse:
         tf.get_variable_scope().reuse_variables()
+      with slim.arg_scope(
+        [slim.batch_norm, slim.dropout], is_training=is_training):  
           with slim.arg_scope(inception.inception_v3_arg_scope()):
             net, _ = inception.inception_v3_base(
                 images, final_endpoint=mparams.final_endpoint)