Exception in thread "main" java.lang.IllegalArgumentException: No tensor type has been registered for data type DT_RESOURCE

[micosacak]

Now I had another problem. I had the following issue (#258) which have been solved. if I use float image;

Exception in thread "main" org.tensorflow.exceptions.TFInvalidArgumentException: Expects arg[0] to be double but float is provided at org.tensorflow.internal.c_api.AbstractTF_Status.throwExceptionIfNotOK(AbstractTF_Status.java:87) at org.tensorflow.Session.run(Session.java:691) at org.tensorflow.Session.access$100(Session.java:72) at org.tensorflow.Session$Runner.runHelper(Session.java:381) at org.tensorflow.Session$Runner.run(Session.java:329) at TfDataTypeIssue.main(TfDataTypeIssue.java:45)

then I converted all buffered image to ´TFloat64´ which is double now I get the following error:

Exception in thread "main" java.lang.IllegalArgumentException: No tensor type has been registered for data type DT_RESOURCE at org.tensorflow.internal.types.registry.TensorTypeRegistry.find(TensorTypeRegistry.java:50) at org.tensorflow.RawTensor.fromHandle(RawTensor.java:147) at org.tensorflow.Session.run(Session.java:695) at org.tensorflow.Session.access$100(Session.java:72) at org.tensorflow.Session$Runner.runHelper(Session.java:381) at org.tensorflow.Session$Runner.run(Session.java:329) at TfDataTypeIssue.main(TfDataTypeIssue.java:45)

here is the part of my Java code;

SavedModelBundle theModel = SavedModelBundle.load("CreateModelWithPythonTensorflow/model", "serve");
double[][][][] floatTensor = getTensoredImage("test_images/test_00001.png");
TFloat64 tensorImage = TFloat64.tensorOf(StdArrays.ndCopyOf(floatTensor));

Tensor res = sess.runner().feed("serving_default_inputTensor_input", tensorImage).fetch("outputTensor/kernel").run().get(0);

or

TFloat64 res = (TFloat64) sess.runner().feed("serving_default_inputTensor_input", tensorImage).fetch("outputTensor/kernel").run().get(0);

[rnett]

Your issue is that outputTensor/kernel is a variable, i.e. a DT_RESOURCE. It's the kernel weight of your last convolution layer. They are essentially handles to storage, they don't have values of their own, so fetching them from a session is meaningless and as such disabled. We could in the future support getting the value of the variable, in the mean time you can add the read op to the graph manually by doing something like this:

Ops newTf = Ops.create(theModel.graph());
Operand<TType> kernelVariable = theModel.graph().operation("outputTensor/kernel").output(0);
Operand<TFloat64> kernel = newTf.readVariableOp(kernelVariable, TFloat64.class);

and then fetch kernel (the operand, not the name) instead of "outputTensor/kernel".

If you're trying to run inference in Java though, this won't work, since you're getting the value of the layer's weight. Looking at your model, the actual output is StatefulPartitionedCall.

[micosacak]

Changing the "outputTensor/kernel" to StatefulPartitionedCall solves the problem. Thanks.