Defines classes to build, save, load and execute TensorFlow models.

WARNING: The API is currently experimental and is not covered by TensorFlow API stability guarantees. See for installation instructions.

The LabelImage example demonstrates use of this API to classify images using a pre-trained Inception architecture convolutional neural network. It demonstrates:

  • Graph construction: using the OperationBuilder class to construct a graph to decode, resize and normalize a JPEG image.
  • Model loading: Using Graph.importGraphDef() to load a pre-trained Inception model.
  • Graph execution: Using a Session to execute the graphs and find the best label for an image.

Additional examples can be found in the tensorflow/models GitHub repository.


ExecutionEnvironment Defines an environment for creating and executing TensorFlow Operations. 
Graph.WhileSubgraphBuilder Used to instantiate an abstract class which overrides the buildSubgraph method to build a conditional or body subgraph for a while loop. 
Operand<T extends TType> Interface implemented by operands of a TensorFlow operation. 
Operation Performs computation on Tensors. 
OperationBuilder A builder for Operations. 
Tensor A statically typed multi-dimensional array. 


ConcreteFunction A graph that can be invoked as a single function, with an input and output signature. 
DeviceSpec Represents a (possibly partial) specification for a TensorFlow device. 
DeviceSpec.Builder A Builder class for building DeviceSpec class. 
EagerSession An environment for executing TensorFlow operations eagerly. 
Graph A data flow graph representing a TensorFlow computation. 
GraphOperation Implementation for an Operation added as a node to a Graph
GraphOperationBuilder An OperationBuilder for adding GraphOperations to a Graph
Output<T extends TType> A symbolic handle to a tensor produced by an Operation
RawTensor A tensor which memory has not been mapped to a data space directly accessible from the JVM. 
SavedModelBundle SavedModelBundle represents a model loaded from storage. 
SavedModelBundle.Exporter Options for exporting a SavedModel. 
SavedModelBundle.Loader Options for loading a SavedModel. 
Server An in-process TensorFlow server, for use in distributed training. 
Session Driver for Graph execution. 
Session.Run Output tensors and metadata obtained when executing a session. 
Session.Runner Run Operations and evaluate Tensors
Signature Describe the inputs and outputs of an executable entity, such as a ConcreteFunction, among other useful metadata. 
Signature.Builder Builds a new function signature. 
TensorFlow Static utility methods describing the TensorFlow runtime. 
TensorMapper<T extends TType> Maps the native memory of a RawTensor to a n-dimensional typed data space accessible from the JVM. 


EagerSession.DevicePlacementPolicy Controls how to act when we try to run an operation on a given device but some input tensors are not on that device.