can't run Tensorflow prediction in Java API

Question

I'm trying to execute a prediction on a model that I trained using "Finetuning AlexNet with TensorFlow" https://kratzert.github.io/2017/02/24/finetuning-alexnet-with-tensorflow.html

I saved the model using tf.saved_model.builder.SavedModelBuilder in Python, and loaded the model in Java using SavedModelBundle.load. the main part of the code is:

    SavedModelBundle smb = SavedModelBundle.load(path, "serve");
    Session s = smb.session();
    byte[] imageBytes = readAllBytesOrExit(Paths.get(path));
    Tensor image = constructAndExecuteGraphToNormalizeImage(imageBytes);
    Tensor result = s.runner().feed("input_tensor", image).fetch("fc8/fc8").run().get(0);
    final long[] rshape = result.shape();
    if (result.numDimensions() != 2 || rshape[0] != 1) {
        throw new RuntimeException(
                String.format(
                        "Expected model to produce a [1 N] shaped tensor where N is the number of labels, instead it produced one with shape %s",
                        Arrays.toString(rshape)));
    }
    int nlabels = (int) rshape[1];
    float [] a =  result.copyTo(new float[1][nlabels])[0];`

I'm getting this Exception:

Exception in thread "main" java.lang.IllegalArgumentException: You must feed a value for placeholder tensor 'Placeholder_1' with dtype float [[Node: Placeholder_1 = Placeholder_output_shapes=[[]], dtype=DT_FLOAT, shape=[], _device="/job:localhost/replica:0/task:0/cpu:0"]]

I saw that the code above worked for some people, and I can't figure out what's missing here. Note that the net is familiar with the nodes "input_tensor" and "fc8/fc8", since it didn't say that it doesn't know them.

ash · Accepted Answer · 2017-06-22 14:38:18Z

1

From the error message, it appears that the model you're using expects to be fed another value (whose node name in the graph is Placeholder_1 and the expected type is a float scalar tensor).

It seems you've customized your model (as opposed to following the article you linked to verbatim). That said, the article shows multiple placeholders that need to be fed, one for the image and another to control dropout. Defined in the article as:

keep_prob = tf.placeholder(tf.float32)

And the value of this placeholder needs to be fed. If you're doing inference, then you want to set keep_prob to 1.0. Something like:

Tensor keep_prob = Tensor.create(1.0f);
Tensor result = s.runner()
  .feed("input_tensor", image)
  .feed("Placeholder_1", keep_prob)
  .fetch("fc8/fc8")
  .run()
  .get(0);

Hope that helps.

edited Jun 22, 2017 at 14:38

answered Jun 20, 2017 at 19:52

ash

6,7714 gold badges20 silver badges30 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

can't run Tensorflow prediction in Java API

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related