org.apache.tika.dl.imagerec

Class DL4JInceptionV3Net

java.lang.Object

org.apache.tika.dl.imagerec.DL4JInceptionV3Net

All Implemented Interfaces:

org.apache.tika.config.Initializable, org.apache.tika.parser.recognition.ObjectRecogniser

public class DL4JInceptionV3Net
extends Object
implements org.apache.tika.parser.recognition.ObjectRecogniser

DL4JInceptionV3Net is an implementation of ObjectRecogniser. This object recogniser is powered by Deeplearning4j. This implementation is pre configured to use Google's InceptionV3 model pre trained on ImageNet corpus. The models references in default settings are originally trained and exported from Keras and imported using DL4J's importer tools.

Although this implementation is made to work out of the box without user attention, for advances users who are interested in tweaking the settings, the following fields are configurable:

• modelWeightsPath
• modelJsonPath
• labelFile
• labelLang
• cacheDir
• imgWidth
• imgHeight
• imgChannels
• minConfidence

Since:

Tika 1.15

See Also:

ObjectRecogniser, ObjectRecognitionParser, org.apache.tika.parser.recognition.tf.TensorflowImageRecParser, org.apache.tika.parser.recognition.tf.TensorflowRESTRecogniser

Field Summary

Fields

Modifier and Type	Field and Description
static String	DEF_LABEL_MAPPING
static String	DEF_MODEL_JSON

Constructor Summary

Constructors

Constructor and Description
DL4JInceptionV3Net()

Method Summary

Modifier and Type	Method and Description
static File	cachedDownload(File cacheDir, URI uri)
void	checkInitialization(org.apache.tika.config.InitializableProblemHandler problemHandler)
Set<org.apache.tika.mime.MediaType>	getSupportedMimes()
void	initialize(Map<String,org.apache.tika.config.Param> params)
boolean	isAvailable()
Map<Integer,String>	loadClassIndex(InputStream stream) Loads the class to
org.nd4j.linalg.api.ndarray.INDArray	preProcessImage(org.nd4j.linalg.api.ndarray.INDArray input) Pre process image to reduce to make it feedable to inception network
List<org.apache.tika.parser.recognition.RecognisedObject>	recognise(InputStream stream, ContentHandler handler, org.apache.tika.metadata.Metadata metadata, org.apache.tika.parser.ParseContext context)

Methods inherited from class java.lang.Object

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

Field Detail

DEF_MODEL_JSON

public static final String DEF_MODEL_JSON

See Also:

Constant Field Values

DEF_LABEL_MAPPING

public static final String DEF_LABEL_MAPPING

See Also:

Constant Field Values

Constructor Detail

DL4JInceptionV3Net

public DL4JInceptionV3Net()

Method Detail

getSupportedMimes

public Set<org.apache.tika.mime.MediaType> getSupportedMimes()

Specified by:

getSupportedMimes in interface org.apache.tika.parser.recognition.ObjectRecogniser

cachedDownload

public static File cachedDownload(File cacheDir,
                                  URI uri)
                           throws IOException

Throws:

IOException

initialize

public void initialize(Map<String,org.apache.tika.config.Param> params)
                throws org.apache.tika.exception.TikaConfigException

Specified by:

initialize in interface org.apache.tika.config.Initializable

Specified by:

initialize in interface org.apache.tika.parser.recognition.ObjectRecogniser

Throws:

org.apache.tika.exception.TikaConfigException

checkInitialization

public void checkInitialization(org.apache.tika.config.InitializableProblemHandler problemHandler)
                         throws org.apache.tika.exception.TikaConfigException

Specified by:

checkInitialization in interface org.apache.tika.config.Initializable

Throws:

org.apache.tika.exception.TikaConfigException

isAvailable

public boolean isAvailable()

Specified by:

isAvailable in interface org.apache.tika.parser.recognition.ObjectRecogniser

preProcessImage

public org.nd4j.linalg.api.ndarray.INDArray preProcessImage(org.nd4j.linalg.api.ndarray.INDArray input)

Pre process image to reduce to make it feedable to inception network

Parameters:

input - Input image

Returns:

processed image

loadClassIndex

public Map<Integer,String> loadClassIndex(InputStream stream)
                                   throws IOException,
                                          org.json.simple.parser.ParseException

Loads the class to

Parameters:

stream - label index stream

Returns:

Map of integer -> label name

Throws:

IOException - when the stream breaks unexpectedly

org.json.simple.parser.ParseException - when the input doesn't contain a valid JSON map

recognise

public List<org.apache.tika.parser.recognition.RecognisedObject> recognise(InputStream stream,
                                                                           ContentHandler handler,
                                                                           org.apache.tika.metadata.Metadata metadata,
                                                                           org.apache.tika.parser.ParseContext context)
                                                                    throws IOException,
                                                                           SAXException,
                                                                           org.apache.tika.exception.TikaException

Specified by:

recognise in interface org.apache.tika.parser.recognition.ObjectRecogniser

Throws:

IOException

SAXException

org.apache.tika.exception.TikaException