Local prediction of ML Engine model

Question

I want to make a local prediction on a saved model which is trained with the ML Engine. For my CNN Model, I'm using the sample code of the flower dataset (https://github.com/GoogleCloudPlatform/cloudml-samples/tree/master/flowers).

I have the following files:

How to restore my model and make a prediction which return a list or dict with the accuracy scores?

I already tried to restore my model but it didn't work out.

import tensorflow as tf

with tf.Session() as sess:    
    saver = tf.train.import_meta_graph('saved_model.pb')
    saver.restore(sess,tf.train.latest_checkpoint('variables/'))
    print(sess.run('w1:0'))

Note 1: My goal is to make a classification report or/and confusion matrix just like we have in Keras.

Note 2: My model consists of 87 labels.

Note 3: Already accessed prediction scores with the Online Prediction API of ML Engine.