TensorFlow Initialise Variables in Variable Scope

In most tutorials the variables initialisation is done with

sess.run(tf.global_variables_initializer())

However if we’re running training of multiple models in a Session, global_variables_initializer() would re-initialise all variables even if they have been trained and set. To initialise variables within a specific scope,

var_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
                             scope=variable_scope)
sess.run(tf.variables_initializer(var_list))