fix bug in f6ede612

tensorpack/models/batch_norm.py

@@ -163,7 +163,12 @@ def BatchNorm(inputs, axis=None, training=None, momentum=0.9, epsilon=1e-5,
            don't want to update it.
         2. As long as `training=True`, `sync_statistics` and `ema_update` option will take effect.
     """
+    # parse training/ctx
     ctx = get_current_tower_context()
+    if training is None:
+        training = ctx.is_training
+    training = bool(training)
+
     # parse shapes
     data_format = get_data_format(data_format, keras_mode=False)
     shape = inputs.get_shape().as_list()
@@ -200,10 +205,6 @@ def BatchNorm(inputs, axis=None, training=None, momentum=0.9, epsilon=1e-5,
 
     TF_version = get_tf_version_tuple()
 
-    # parse training/ctx
-    if training is None:
-        training = ctx.is_training
-    training = bool(training)
     freeze_bn_backward = not training and ctx.is_training
     if freeze_bn_backward:
         assert TF_version >= (1, 4), \
@@ -212,6 +213,7 @@ def BatchNorm(inputs, axis=None, training=None, momentum=0.9, epsilon=1e-5,
             logger.warn("[BatchNorm] Using moving_mean/moving_variance in training.")
         # Using moving_mean/moving_variance in training, which means we
         # loaded a pre-trained BN and only fine-tuning the affine part.
+
     do_sync_bn = (sync_statistics is not None) and training
 
     if not do_sync_bn: