axip.ailia

Class Ailia

java.lang.Object

axip.ailia.Ailia

public class Ailia
extends java.lang.Object

AILIA inference library (native API)

Field Summary

Fields

Modifier and Type	Field and Description
static int	AILIA_MEMORY_REDUCE_CONSTANT_WITH_FILE_MAPPED Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT_WITH_FILE_MAPPED
static int	ENVIRONMENT_ID_AUTO Automatic setup of the inference backend
static int	MEMORY_NO_OPTIMIZATION Deprecated. Please use AiliaMemoryMode.NO_OPTIMIZATION
static int	MEMORY_OPTIMAIZE_DEFAULT Deprecated. Please use AiliaMemoryMode.OPTIMAIZE_DEFAULT
static int	MEMORY_REDUCE_CONSTANT Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT
static int	MEMORY_REDUCE_CONSTANT_WITH_INPUT_INITIALIZER Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT_WITH_INPUT_INITIALIZER
static int	MEMORY_REDUCE_INTERSTAGE Deprecated. Please use AiliaMemoryMode.REDUCE_INTERSTAGE
static int	MEMORY_REUSE_INTERSTAGE Deprecated. Please use AiliaMemoryMode.REUSE_INTERSTAGE
static int	MULTITHREAD_AUTO Number of threads

Constructor Summary

Constructors

Constructor and Description
Ailia()

Method Summary

Modifier and Type	Method and Description
static void	CopyBlobData(long dstHandle, int dstBlobIdx, long srcHandle, int srcBlobIdx) Perform copies between blobs that specified.
static long	Create(int envId, int numThread) Creates a network instance.
static void	Destroy(long handle) It destroys the network instance.
static void	DisableLayerFusion(long handle) Disalbe layer fusion optimaization for inference
static void	Finalize() Release GPU and other resources that have been globally allocated with ailia.
static int	FindBlobIndexByName(long handle, java.lang.String name) Searches by name for the index of the internal data (blob) during inference and returns it.
static java.lang.String	FindBlobNameByIndex(long handle, int blobIdx) Searches by index for the name of the internal data (blob) during inference and returns it.
static int	GetBlobCount(long handle) Gets the amount of internal data (blob) during inference.
static void	GetBlobData(long handle, float[] dest, int destSize, int blobIdx) Gets the internal data (blob) during inference.
static int	GetBlobDim(long handle, int blobIdx) Gets the dimension of the internal data (blob) during inference.
static int	GetBlobIndexByInputIndex(long handle, int inputBlobIdx) Get the blob index of the input data.
static int	GetBlobIndexByOutputIndex(long handle, int outputBlobIdx) Get the blob index of the output data blob.
static AiliaShape	GetBlobShape(long handle, int blobIdx, int version) Gets the shape of the internal data (blob) during inference.
static void	GetBlobShapeND(long handle, int[] shape, int dim, int blobIdx) Gets the shape of internal data (blob) during inference.
static AiliaEnvironment	GetEnvironment(int env_id, int version) Gets the list of computational environments.
static int	GetEnvironmentCount() Gets the number of available computational environments (CPU, GPU).
static java.lang.String	GetErrorDetail(long handle) Returns the details of errors.
static int	GetInputBlobCount(long handle) Get the number of input data blobs.
static int	GetInputDim(long handle) Gets the dimension of the input data during inference.
static AiliaShape	GetInputShape(long handle, int version) Gets the shape of the input data during inference.
static void	GetInputShapeND(long handle, int[] shape, int dim) Gets the shape of the input data during inference.
static int	GetOutputBlobCount(long handle) Get the number of output data blobs.
static int	GetOutputDim(long handle) Gets the dimension of the output data during inference.
static AiliaShape	GetOutputShape(long handle, int version) Gets the shape of the output data during inference.
static void	GetOutputShapeND(long handle, int[] shape, int dim) Gets the shape of the output data during inference.
static AiliaEnvironment	GetSelectedEnvironment(long handle, int version) Gets the selected computational environment.
static java.lang.String	GetVersion() Get the version of the library.
static void	OpenStreamFile(long handle, java.lang.String path) Initializes the network instance.
static void	OpenStreamMem(long handle, byte[] buf, int buf_size) Initializes the network instance.
static void	OpenWeightFile(long handle, java.lang.String path) Reads weights into a network instance.
static void	OpenWeightMem(long handle, byte[] buf, int buf_size) Reads weights into a network instance.
static void	Predict(long handle, float[] dest, int destSize, float[] src, int srcSize) Performs the inferences and provides the inference result.
static void	SetInputBlobData(long handle, float[] src, int srcSize, int blobIdx) Provides the specified blob with the input data.
static void	SetInputBlobShape(long handle, AiliaShape shape, int blobIdx, int version) Change the shape of the blob given by its index
static void	SetInputBlobShapeND(long handle, int[] shape, int dim, int blobIdx) Change the shape of the blob given by its index
static void	SetInputShape(long handle, AiliaShape shape, int version) Changes the shape of the input data during inference.
static void	SetInputShapeND(long handle, int[] shape, int dim) Changes the shape of the input data during inference.
static void	SetMemoryMode(long handle, int mode) Set the memory usage policy for inference
static void	SetProfileMode(long handle, int mode) Set the profile mode.
static void	SetTemporaryCachePath(java.lang.String cache_path) Specifies a temporary cache directory.
static java.lang.String	Summary(long handle) Shows the name and shape of each blob.
static void	Update(long handle) Makes inferences with the input data specified in advance.

Methods inherited from class java.lang.Object

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

Field Detail

MULTITHREAD_AUTO

public static final int MULTITHREAD_AUTO

Number of threads

See Also:

Constant Field Values

ENVIRONMENT_ID_AUTO

public static final int ENVIRONMENT_ID_AUTO

Automatic setup of the inference backend

See Also:

Constant Field Values

MEMORY_NO_OPTIMIZATION

@Deprecated
public static final int MEMORY_NO_OPTIMIZATION

Deprecated. Please use AiliaMemoryMode.NO_OPTIMIZATION

Do not release the intermediate buffer

See Also:

Constant Field Values

MEMORY_REDUCE_CONSTANT

@Deprecated
public static final int MEMORY_REDUCE_CONSTANT

Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT

Releases the intermediate buffer that is a constant such as weight

See Also:

Constant Field Values

MEMORY_REDUCE_CONSTANT_WITH_INPUT_INITIALIZER

@Deprecated
public static final int MEMORY_REDUCE_CONSTANT_WITH_INPUT_INITIALIZER

Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT_WITH_INPUT_INITIALIZER

Disable the input specified initializer and release the intermediate buffer that becomes a constant such as weight.

See Also:

Constant Field Values

MEMORY_REDUCE_INTERSTAGE

@Deprecated
public static final int MEMORY_REDUCE_INTERSTAGE

Deprecated. Please use AiliaMemoryMode.REDUCE_INTERSTAGE

Release intermediate buffer during inference

See Also:

Constant Field Values

MEMORY_REUSE_INTERSTAGE

@Deprecated
public static final int MEMORY_REUSE_INTERSTAGE

Deprecated. Please use AiliaMemoryMode.REUSE_INTERSTAGE

Infer by sharing the intermediate buffer. When used with MEMORY_REDUCE_INTERSTAGE, the sharable intermediate buffer is not opened.

See Also:

Constant Field Values

AILIA_MEMORY_REDUCE_CONSTANT_WITH_FILE_MAPPED

@Deprecated
public static final int AILIA_MEMORY_REDUCE_CONSTANT_WITH_FILE_MAPPED

Deprecated. Please use AiliaMemoryMode.REDUCE_CONSTANT_WITH_FILE_MAPPED

Use memory mapped file due to reduce constant intermediate blobs. Need to call SetTemporaryCachePath in advance.

See Also:

Constant Field Values

MEMORY_OPTIMAIZE_DEFAULT

@Deprecated
public static final int MEMORY_OPTIMAIZE_DEFAULT

Deprecated. Please use AiliaMemoryMode.OPTIMAIZE_DEFAULT

See Also:

Constant Field Values

Constructor Detail

Ailia

public Ailia()

Method Detail

Create

public static long Create(int envId,
                          int numThread)
                   throws AiliaException

Creates a network instance.

If the inference backend is set to automatic, CPU mode is used, while if BLAS is available, it uses BLAS. Note that if BLAS is used, num_thread may be ignored.

Parameters:

envId - The ID of the inference backend used for computation (obtained by GetEnvironment(int, int)). It is selected automatically if ENVIRONMENT_ID_AUTO is specified.

numThread - The upper limit on the number of threads (It is set automatically if MULTITHREAD_AUTO is specified.)

Returns:

Handle of network instance.

Throws:

AiliaException - Exception

OpenStreamFile

public static void OpenStreamFile(long handle,
                                  java.lang.String path)
                           throws AiliaException

Initializes the network instance. (Read from file)

This function reads the network instance from a file and initializes it.

Parameters:

handle - A network instance handle

path - The path name to the prototxt file

Throws:

AiliaException - Exception

OpenStreamMem

public static void OpenStreamMem(long handle,
                                 byte[] buf,
                                 int buf_size)
                          throws AiliaException

Initializes the network instance. (Read from memory)

This function reads the network instance from memory and initializes it.

Parameters:

handle - A network instance handle

buf - An array of the data in the protobuf file

buf_size - The data size of the prototxt file

Throws:

AiliaException - Exception

OpenWeightFile

public static void OpenWeightFile(long handle,
                                  java.lang.String path)
                           throws AiliaException

Reads weights into a network instance. (Read from file)

This function reads weights into the network instance from a file.

Parameters:

handle - A network instance handle

path - The path name to the protobuf/onnx file

Throws:

AiliaException - Exception

OpenWeightMem

public static void OpenWeightMem(long handle,
                                 byte[] buf,
                                 int buf_size)
                          throws AiliaException

Reads weights into a network instance. (Read from memory) This function reads weights into the network instance from memory.

Parameters:

handle - A network instance handle

buf - An array of the data in the protobuf/onnx file

buf_size - The data size of the protobuf/onnx file

Throws:

AiliaException - Exception

Destroy

public static void Destroy(long handle)

It destroys the network instance.

Parameters:

handle - A network instance handle

SetInputShape

public static void SetInputShape(long handle,
                                 AiliaShape shape,
                                 int version)
                          throws AiliaException

Changes the shape of the input data during inference.

This function changes the input shape defined in prototxt. The shape must have the same rank as the one contained in prototxt. Note that AiliaException may be throw if the weights are dependent on the input shapes, among other reasons. The dimension of shape that defined in prototxt has 5 or more, please use SetInputShapeND(long, int[], int).

Parameters:

handle - A network instance handle

shape - Shape information for the input data

version - Shape version(AiliaShape.version)

Throws:

AiliaException - Exception

SetInputShapeND

public static void SetInputShapeND(long handle,
                                   int[] shape,
                                   int dim)
                            throws AiliaException

Changes the shape of the input data during inference.

This function changes the input shape defined in prototxt. The shape must have the same rank as the one contained in prototxt. Note that AiliaException may be throw if the weights are dependent on the input shapes, among other reasons.

Parameters:

handle - A network instance handle

shape - An array of shape

dim - The size of shape.

Throws:

AiliaException - Exception

GetInputShape

public static AiliaShape GetInputShape(long handle,
                                       int version)
                                throws AiliaException

Gets the shape of the input data during inference.

When dimension of shape is 5 or more, please use GetInputDim(long) and GetInputShapeND(long, int[], int). When shape is not settled, this method stores 0 at unsettled dimension and otherwise stores valid value.

Parameters:

handle - A network instance handle

version - Shape version(AiliaShape.version)

Returns:

The shape of the input data

Throws:

AiliaException - Exception

GetInputDim

public static int GetInputDim(long handle)

Gets the dimension of the input data during inference.

Parameters:

handle - A network instance handle

Returns:

The dimension of the input data.

GetInputShapeND

public static void GetInputShapeND(long handle,
                                   int[] shape,
                                   int dim)

Gets the shape of the input data during inference.

When shape is not settled, this function stores 0 at unsettled dimension and otherwise stores valid value.

Parameters:

handle - A network instance handle

shape - The storage array of the shape. (It stores dim-1, dim-2, ... ,1, 0 order.)

dim - The size of shape

GetOutputShape

public static AiliaShape GetOutputShape(long handle,
                                        int version)
                                 throws AiliaException

Gets the shape of the output data during inference.

When dimension of shape is 5 or more, please use GetOutputDim(long) and GetOutputShapeND(long, int[], int).

Parameters:

handle - A network instance handle

version - Shape version(AiliaShape.version)

Returns:

The shape of the output.

Throws:

AiliaException - Exception

GetOutputDim

public static int GetOutputDim(long handle)

Gets the dimension of the output data during inference.

Parameters:

handle - A network instance handle

Returns:

The dimension of the output.

GetOutputShapeND

public static void GetOutputShapeND(long handle,
                                    int[] shape,
                                    int dim)

Gets the shape of the output data during inference.

Parameters:

handle - A network instance handle

shape - The storage array of the shape. (It stores dim-1, dim-2, ... ,1, 0 order.)

dim - The size of shape

Predict

public static void Predict(long handle,
                           float[] dest,
                           int destSize,
                           float[] src,
                           int srcSize)
                    throws AiliaException

Performs the inferences and provides the inference result.

Parameters:

handle - A network instance handle

dest - The result is stored in the inference result destination buffer as float data in the order of X, Y, Z, and W. The buffer has the same size as the network file outputSize.

destSize - The size of the dest (in byte)

src - The input is stored as float data in the order of the inference data X, Y, Z, and W. The input has the same size as the network file inputSize.

srcSize - The size of the src (in byte)

Throws:

AiliaException - Exception

GetBlobCount

public static int GetBlobCount(long handle)
                        throws AiliaException

Gets the amount of internal data (blob) during inference.

Parameters:

handle - A network instance handle

Returns:

The number of blobs

Throws:

AiliaException - Exception

GetBlobShape

public static AiliaShape GetBlobShape(long handle,
                                      int blobIdx,
                                      int version)
                               throws AiliaException

Gets the shape of the internal data (blob) during inference.

Parameters:

handle - A network instance handle

blobIdx - The index of the blob (0 to GetBlobCount(long)-1)

version - Shape version(AiliaShape.version)

Returns:

The shape of the internal data (blob).

Throws:

AiliaException - Exception

GetBlobDim

public static int GetBlobDim(long handle,
                             int blobIdx)

Gets the dimension of the internal data (blob) during inference.

Parameters:

handle - A network instance handle

blobIdx - The index of the blob (0 to GetBlobCount(long)-1)

Returns:

The dimension of the internal data (blob).

GetBlobShapeND

public static void GetBlobShapeND(long handle,
                                  int[] shape,
                                  int dim,
                                  int blobIdx)

Gets the shape of internal data (blob) during inference.

Parameters:

handle - A network instance handle

shape - The storage array of the shape. (It stores dim-1, dim-2, ... ,1, 0 order.)

dim - The size of shape

blobIdx - The index of the blob (0 to GetBlobCount(long)-1)

GetBlobData

public static void GetBlobData(long handle,
                               float[] dest,
                               int destSize,
                               int blobIdx)
                        throws AiliaException

Gets the internal data (blob) during inference.

Parameters:

handle - A network instance handle

dest - The result is stored in the inference result destination buffer as float data in the order of X, Y, Z, and W.

destSize - The size of the dest (in byte)

blobIdx - The index of the blob (0 to GetBlobCount(long)-1)

Throws:

AiliaException - Exception

FindBlobIndexByName

public static int FindBlobIndexByName(long handle,
                                      java.lang.String name)
                               throws AiliaException

Searches by name for the index of the internal data (blob) during inference and returns it.

Parameters:

handle - A network instance handle

name - The name of the blob to search for

Returns:

The index of the blob (0 to GetBlobCount(long)-1)

Throws:

AiliaException - Exception

FindBlobNameByIndex

public static java.lang.String FindBlobNameByIndex(long handle,
                                                   int blobIdx)
                                            throws AiliaException

Searches by index for the name of the internal data (blob) during inference and returns it.

Parameters:

handle - A network instance handle

blobIdx - The index of the blob (0 to GetBlobCount(long)-1)

Returns:

The blob name

Throws:

AiliaException - Exception

Summary

public static java.lang.String Summary(long handle)
                                throws AiliaException

Shows the name and shape of each blob.

Parameters:

handle - A network instance handle

Returns:

The summary string

Throws:

AiliaException - Exception

GetInputBlobCount

public static int GetInputBlobCount(long handle)
                             throws AiliaException

Get the number of input data blobs.

Parameters:

handle - A network instance handle

Returns:

The number of input blobs

Throws:

AiliaException - Exception

GetBlobIndexByInputIndex

public static int GetBlobIndexByInputIndex(long handle,
                                           int inputBlobIdx)
                                    throws AiliaException

Get the blob index of the input data.

Parameters:

handle - A network instance handle

inputBlobIdx - Index among the input blobs (between 0 and GetInputBlobCount(long)-1)

Returns:

Index of the blob (between 0 and GetBlobCount(long)-1)

Throws:

AiliaException - Exception

SetInputBlobData

public static void SetInputBlobData(long handle,
                                    float[] src,
                                    int srcSize,
                                    int blobIdx)
                             throws AiliaException

Provides the specified blob with the input data.

This method is used to specify the input on networks with multiple inputs.

Parameters:

handle - A network instance handle

src - The inference data is stored as float data in the order of X, Y, Z, and W.

srcSize - The size of the inference data (in byte)

blobIdx - The index of the blob for input

Throws:

AiliaException - Exception

SetInputBlobShape

public static void SetInputBlobShape(long handle,
                                     AiliaShape shape,
                                     int blobIdx,
                                     int version)
                              throws AiliaException

Change the shape of the blob given by its index

This is useful to change the network input shape in a context where there are several input blobs. If dimension of shape has 5 or more, please use SetInputBlobShapeND(long, int[], int, int)

Parameters:

handle - A network instance handle

shape - New shape of the blob

blobIdx - Index referencing the blob to reshape

version - Shape version(AiliaShape.version)

Throws:

AiliaException - Exception

See Also:

SetInputShape(long, AiliaShape, int)

SetInputBlobShapeND

public static void SetInputBlobShapeND(long handle,
                                       int[] shape,
                                       int dim,
                                       int blobIdx)
                                throws AiliaException

Change the shape of the blob given by its index

This is useful to change the network input shape in a context where there are several input blobs.

Parameters:

handle - A network instance handle

shape - The array of the new shape. (It stores dim-1, dim-2, ... ,1, 0 order.)

dim - The size of shape

blobIdx - Index referencing the blob to reshape

Throws:

AiliaException - Exception

See Also:

SetInputShapeND(long, int[], int)

CopyBlobData

public static void CopyBlobData(long dstHandle,
                                int dstBlobIdx,
                                long srcHandle,
                                int srcBlobIdx)
                         throws AiliaException

Perform copies between blobs that specified.

Parameters:

dstHandle - A network instance handle that contains destination blob

dstBlobIdx - A network instance handle The index of destination blob

srcHandle - A network instance handle that contains source blob

srcBlobIdx - A network instance handle The index of source blob

Throws:

AiliaException - Exception

Update

public static void Update(long handle)
                   throws AiliaException

Makes inferences with the input data specified in advance.

This function is used when, for example, the input is provided with SetInputBlobData(long, float[], int, int). Get the inference result with GetBlobData(long, float[], int, int)

Parameters:

handle - A network instance handle

Throws:

AiliaException - Exception

GetOutputBlobCount

public static int GetOutputBlobCount(long handle)
                              throws AiliaException

Get the number of output data blobs.

Parameters:

handle - A network instance handle

Returns:

The number of output blobs

Throws:

AiliaException - Exception

GetBlobIndexByOutputIndex

public static int GetBlobIndexByOutputIndex(long handle,
                                            int outputBlobIdx)
                                     throws AiliaException

Get the blob index of the output data blob.

Parameters:

handle - A network instance handle

outputBlobIdx - Index among output blobs (between 0 and GetOutputBlobCount(long)-1)

Returns:

Blob index (between 0 and GetBlobCount(long)-1)

Throws:

AiliaException - Exception

SetTemporaryCachePath

public static void SetTemporaryCachePath(java.lang.String cache_path)
                                  throws AiliaException

Specifies a temporary cache directory.

This system uses the specified cache directory to generate and store machine code optimized for each inference backend. Call only once at the start of execution of ailia. It ignores any second and subsequent calls, and automatically returns success. There is no particular problem if it is called from multiple threads, as it provides exclusive control internally. Some functions, such as Vulkan shader cache, cannot be used until this API function is called. Specify the file path obtained by android.content.Context#getCacheDir() for cache_dir.

Parameters:

cache_path - Temporary cache directory

Throws:

AiliaException - Exception

GetEnvironmentCount

public static int GetEnvironmentCount()
                               throws AiliaException

Gets the number of available computational environments (CPU, GPU).

Returns:

The number of computational environment information

Throws:

AiliaException - Exception

GetEnvironment

public static AiliaEnvironment GetEnvironment(int env_id,
                                              int version)
                                       throws AiliaException

Gets the list of computational environments.

Parameters:

env_id - The index of the computational environment information (0 to GetEnvironmentCount()-1)

version - AiliaEnvironment.version

Returns:

The computational environment information (valid until the AILIANetwork instance is destroyed)

Throws:

AiliaException - Exception

GetSelectedEnvironment

public static AiliaEnvironment GetSelectedEnvironment(long handle,
                                                      int version)
                                               throws AiliaException

Gets the selected computational environment.

Parameters:

handle - A network instance handle

version - AiliaEnvironment.version

Returns:

The computational environment information (valid until the AILIANetwork instance is destroyed)

Throws:

AiliaException - Exception

SetMemoryMode

public static void SetMemoryMode(long handle,
                                 int mode)
                          throws AiliaException

Set the memory usage policy for inference

Change the memory usage policy. If a value other than MEMORY_NO_OPTIMIZATION is specified, the intermediate buffer secured during inference will be released, so the memory usage during inference can be reduced. Must be specified immediately after ailiaCreate. It cannot be changed after calling ailiaOpen. If you specify to release the intermediate buffer, calling GetBlobData(long, float[], int, int) for the corresponding blob will thrown AiliaException.

Parameters:

handle - A network instance handle

mode - Memory mode (Multiple specifications possible with logical sum) AiliaMemoryMode (Default :AiliaMemoryMode.OPTIMAIZE_DEFAULT)

Throws:

AiliaException - Exception

DisableLayerFusion

public static void DisableLayerFusion(long handle)
                               throws AiliaException

Disalbe layer fusion optimaization for inference

This api use to get blob that remove by layer fusion optimization. Must be specified immediately after ailiaCreate. It cannot be changed after calling ailiaOpen. Note: When disable layer fusion optimization, inference speed may be down.

Parameters:

handle - A network instance handle

Throws:

AiliaException

SetProfileMode

public static void SetProfileMode(long handle,
                                  int mode)
                           throws AiliaException

Set the profile mode.

Change the profile mode. If a value other than AiliaProfileMode.PROFILE_DISABLE is specified, the profile data is added to Summary

Parameters:

handle - A network instance handle

mode - Profile mode (Default :AiliaProfileMode.PROFILE_DISABLE)

Throws:

AiliaException - Exception

GetErrorDetail

public static java.lang.String GetErrorDetail(long handle)

Returns the details of errors.

Parameters:

handle - A network instance handle

Returns:

Error details

GetVersion

public static java.lang.String GetVersion()

Get the version of the library.

Returns:

Version number

Finalize

public static void Finalize()
                     throws AiliaException

Release GPU and other resources that have been globally allocated with ailia.

Throws:

AiliaException - Exception