org.tensorflow.contrib.android

Class TensorFlowInferenceInterface

java.lang.Object

org.tensorflow.contrib.android.TensorFlowInferenceInterface

public class TensorFlowInferenceInterface
extends Object

Wrapper over the TensorFlow API (Graph, Session) providing a smaller API surface for inference.

See tensorflow/examples/android/src/org/tensorflow/demo/TensorFlowImageClassifier.java for an example usage.

Constructor Summary

Constructors

Constructor and Description
TensorFlowInferenceInterface(android.content.res.AssetManager assetManager, String model)
TensorFlowInferenceInterface(Graph g)
TensorFlowInferenceInterface(InputStream is)

Method Summary

Modifier and Type	Method and Description
void	close() Cleans up the state associated with this Object.
void	feed(String inputName, boolean[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, byte[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, ByteBuffer src, long... dims) Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, double[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, DoubleBuffer src, long... dims) Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, float[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, FloatBuffer src, long... dims) Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, int[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, IntBuffer src, long... dims) Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, long[] src, long... dims) Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName.
void	feed(String inputName, LongBuffer src, long... dims) Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName.
void	feedString(String inputName, byte[] src) Copy a byte sequence into the input Tensor with name inputName as a string-valued scalar tensor.
void	feedString(String inputName, byte[][] src) Copy an array of byte sequences into the input Tensor with name inputName as a string-valued one-dimensional tensor (vector).
void	fetch(String outputName, byte[] dst) Read from a Tensor named outputName and copy the contents into a Java array.
void	fetch(String outputName, ByteBuffer dst) Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst.
void	fetch(String outputName, double[] dst) Read from a Tensor named outputName and copy the contents into a Java array.
void	fetch(String outputName, DoubleBuffer dst) Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst.
void	fetch(String outputName, float[] dst) Read from a Tensor named outputName and copy the contents into a Java array.
void	fetch(String outputName, FloatBuffer dst) Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst.
void	fetch(String outputName, int[] dst) Read from a Tensor named outputName and copy the contents into a Java array.
void	fetch(String outputName, IntBuffer dst) Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst.
void	fetch(String outputName, long[] dst) Read from a Tensor named outputName and copy the contents into a Java array.
void	fetch(String outputName, LongBuffer dst) Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst.
protected void	finalize()
String	getStatString() Returns the last stat summary string if logging is enabled.
Graph	graph() Returns a reference to the Graph describing the computation run during inference.
Operation	graphOperation(String operationName)
void	run(String[] outputNames) Runs inference between the previously registered input nodes (via feed*) and the requested output nodes.
void	run(String[] outputNames, boolean enableStats) Runs inference between the previously registered input nodes (via feed*) and the requested output nodes.
void	run(String[] outputNames, boolean enableStats, String[] targetNodeNames) An overloaded version of runInference that allows supplying targetNodeNames as well

Methods inherited from class java.lang.Object

clone, equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TensorFlowInferenceInterface

public TensorFlowInferenceInterface(android.content.res.AssetManager assetManager,
                                    String model)

TensorFlowInferenceInterface

public TensorFlowInferenceInterface(InputStream is)

TensorFlowInferenceInterface

public TensorFlowInferenceInterface(Graph g)

Method Detail

run

public void run(String[] outputNames)

Runs inference between the previously registered input nodes (via feed*) and the requested output nodes. Output nodes can then be queried with the fetch* methods.

Parameters:

outputNames - A list of output nodes which should be filled by the inference pass.

run

public void run(String[] outputNames,
                boolean enableStats)

Runs inference between the previously registered input nodes (via feed*) and the requested output nodes. Output nodes can then be queried with the fetch* methods.

Parameters:

outputNames - A list of output nodes which should be filled by the inference pass.

run

public void run(String[] outputNames,
                boolean enableStats,
                String[] targetNodeNames)

An overloaded version of runInference that allows supplying targetNodeNames as well

graph

public Graph graph()

Returns a reference to the Graph describing the computation run during inference.

graphOperation

public Operation graphOperation(String operationName)

getStatString

public String getStatString()

Returns the last stat summary string if logging is enabled.

close

public void close()

Cleans up the state associated with this Object.

The TenosrFlowInferenceInterface object is no longer usable after this method returns.

finalize

protected void finalize()
                 throws Throwable

Overrides:

finalize in class Object

Throws:

Throwable

feed

public void feed(String inputName,
                 boolean[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 float[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 int[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 long[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 double[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 byte[] src,
                 long... dims)

Given a source array with shape dims and content src, copy the contents into the input Tensor with name inputName. The source array src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feedString

public void feedString(String inputName,
                       byte[] src)

Copy a byte sequence into the input Tensor with name inputName as a string-valued scalar tensor. In the TensorFlow type system, a "string" is an arbitrary sequence of bytes, not a Java String (which is a sequence of characters).

feedString

public void feedString(String inputName,
                       byte[][] src)

Copy an array of byte sequences into the input Tensor with name inputName as a string-valued one-dimensional tensor (vector). In the TensorFlow type system, a "string" is an arbitrary sequence of bytes, not a Java String (which is a sequence of characters).

feed

public void feed(String inputName,
                 FloatBuffer src,
                 long... dims)

Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName. The source buffer src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 IntBuffer src,
                 long... dims)

Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName. The source buffer src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 LongBuffer src,
                 long... dims)

Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName. The source buffer src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 DoubleBuffer src,
                 long... dims)

Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName. The source buffer src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

feed

public void feed(String inputName,
                 ByteBuffer src,
                 long... dims)

Given a source buffer with shape dims and content src, both stored as direct and native ordered java.nio buffers, copy the contents into the input Tensor with name inputName. The source buffer src must have at least as many elements as that of the destination Tensor. If src has more elements than the destination has capacity, the copy is truncated.

fetch

public void fetch(String outputName,
                  float[] dst)

Read from a Tensor named outputName and copy the contents into a Java array. dst must have length greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  int[] dst)

Read from a Tensor named outputName and copy the contents into a Java array. dst must have length greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  long[] dst)

Read from a Tensor named outputName and copy the contents into a Java array. dst must have length greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  double[] dst)

Read from a Tensor named outputName and copy the contents into a Java array. dst must have length greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  byte[] dst)

Read from a Tensor named outputName and copy the contents into a Java array. dst must have length greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  FloatBuffer dst)

Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst. dst must have capacity greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  IntBuffer dst)

Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst. dst must have capacity greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  LongBuffer dst)

Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst. dst must have capacity greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  DoubleBuffer dst)

Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst. dst must have capacity greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.

fetch

public void fetch(String outputName,
                  ByteBuffer dst)

Read from a Tensor named outputName and copy the contents into the direct and native ordered java.nio buffer dst. dst must have capacity greater than or equal to that of the source Tensor. This operation will not affect dst's content past the source Tensor's size.