In tf.slim, whether I need to add the dependency to the loss

Question

In tf.slim, I have used the batch_norm.

My question is: whether I need explicitly to add the dependency to the loss?

I think, the slim knew I have used the batch_norm, whether it has automatically add the dependency to the loss? I am very confused.

Answer 1

Yes, you need.

Could you follow the instructions here:

Note: when training, the moving_mean and moving_variance need to be updated. By default the update ops are placed in tf.GraphKeys.UPDATE_OPS, so they need to be added as a dependency to the train_op. For example:

update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
  train_op = optimizer.minimize(loss)