CODEWITHSUNDEEP
tensorflowrestorepre-trained-model
user1948893
user1948893

Reputation: 73

How to restore a tensorflow model that only has one file with extension ".model"

I want to use a pretrained tensorflow model provided by an unknown author. I do not know how he/she managed to save the tensorflow model (he/she used tensorflow version >= 1.2) to only one file with the extension '.model', as normally I get either three files '.meta', '.data', '.index' or one file with '.ckpt'.

How can I restore this pretrained model? How can I save a model to this format later?

Thanks.

Upvotes: 3

Views: 1252

Answers (1)

rocksyne
rocksyne

Reputation: 1372

I have also asked this question on a number of platforms with no assistance yet. So I decided to do some experimental work and this is what I found. This may be long but please bear with me.

To import a model in Tensor-flow we use

with tf.Session() as sess:
  new_saver = tf.train.import_meta_graph('my_test_model-1000.meta')
  new_saver.restore(sess, tf.train.latest_checkpoint('./'))

The .meta file contains all the variables, operations, collections, etc, of the trained model. What tf.train.latest_checkpoint('./') does is to use the checkpoint file (which simply keeps a record of latest checkpoint files saved) to import the xxxx_model.data-00000-of-00001. This .data-00000-of-00001 contains all the weights, biases, gradients, etc, that must be loaded into the variables contained in my_test_model-1000.meta.

Summary [Semi-complete code]

with tf.Session() as sess:
    new_saver = tf.train.import_meta_graph('my_test_model-1000.meta')
    #new_saver.restore(sess, tf.train.latest_checkpoint('./'))
    tensor_variable = tf.trainable_variables()
    for tensor_var in tensor_variable:
        #print(sess.run(tensor_var))
        print(tensor_var)

This initial code will print out all the variables from .meta that are trainable. If you try to run print(sess.run(tensor_var)) you will get an error. This is because, the variables have not been initialized. How ever, if you un-comment new_saver.restore(sess, tf.train.latest_checkpoint('./')) and run print(sess.run(tensor_var)), you will get all the variables alongside values loaded into the variables.

Now to “.model”

My best guess is that xxxxxx.model works a much like xxxx_model.data-00000-of-00001 from tensorflow. It does not contain variables and so if you try to do 

with tf.Session() as sess:
          new_saver = tf.train.import_meta_graph('xxx.model')

you will get an error. Remember, the reason is that, this .model file does not contain any variables nor operation graph of any form. If you also try to do 

with tf.Session() as sess:
          new_saver = tf.train.Saver()
          new_saver.restore(sess, "xxxx.model")

you will similarly get an error. This is because, there are no corresponding variables to load values into. Therefore, if you ever obtain a xxx.model file, you will have to go through the pain of replicating all the variables and operations before trying to run new_saver.restore(sess, "xxxx.model"). If you are able to replicate the architecture, this will run smoothly with no issues, hopefully.

I am sorry this was long, but considering that there is almost no answer on the internet, I had to make a lecture out of it. :)

Upvotes: 2

Related Questions