Minor issue with name scopes

[ameroyer]

Hello,
I've noticed a minor issue when using name scopes (mostly for Tensorboard visualization). For instance the following code raises an error in the utils.var_scope wrapper:

with tf.name_scope('A'):
        with tf.variable_scope('B'):
            x = tf.placeholder(shape=(None, 1024, 1024, 3), dtype=tf.float32)
            y = tensornets.MobileNet100(x, stem=True)

It seems the problem comes from line 243 where _scope = tf.get_variable_scope().name only looks at variable scope while the tf.get_collection(..., scope=_scope) at line 124 look through variable names.
It can be solved by replacing line 243 with _scope = tf.contrib.framework.get_name_scope(), but I'm not sure if it's the best workaround.

[taehoonlee]

Thank you for the report, @ameroyer.

1. Fix: ae8e431
2. Notes: Notes for variable_scope, name_scope, and weight sharing #43
3. Test codes: b1c2533

If you want to use the latest version, you can try pip install -U tensornets. Please feel free to open it again at any time if you have additional comments :)