bump TF required version

README.md

@@ -62,7 +62,7 @@ You can also define your own trainer for non-standard training (e.g. GAN).
 Dependencies:
 
 + Python 2 or 3
-+ TensorFlow >= 1.0.0rc1
++ TensorFlow >= 1.0.0
 + Python bindings for OpenCV
 ```
 pip install --user -U git+https://github.com/ppwwyyxx/tensorpack.git

tensorpack/models/batch_norm.py

@@ -124,6 +124,8 @@ def update_bn_ema(xn, batch_mean, batch_var, moving_mean, moving_var, decay):
         name='var_ema_op')
     add_model_variable(moving_mean)
     add_model_variable(moving_var)
+
+    # seems faster than delayed update, but might behave otherwise in distributed settings.
     with tf.control_dependencies([update_op1, update_op2]):
         return tf.identity(xn, name='output')