org.openimaj.image.indexing.vlad

Class VLADIndexerData

java.lang.Object

org.openimaj.image.indexing.vlad.VLADIndexerData

public class VLADIndexerData
extends Object

Class representing the data required to build a VLAD + PCA + product-quantisation based image index.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Constructor Summary

Constructors

Constructor and Description
VLADIndexerData(VLAD<float[]> vlad, FeatureVectorPCA pca, FloatProductQuantiser pq, LocalFeatureExtractor<LocalFeature<?,?>,MBFImage> extractor, Function<List<? extends LocalFeature<?,?>>,List<FloatLocalFeatureAdaptor<?>>> postProcess) Construct with the given data

Method Summary

Modifier and Type	Method and Description
IncrementalFloatADCNearestNeighbours	createIncrementalIndex() Create an IncrementalFloatADCNearestNeighbours pre-prepared to index data
float[]	extractPcaVlad(List<? extends LocalFeature<?,?>> features) Extract the PCA-projected VLAD feature from the given raw local features.
float[]	extractPcaVlad(MBFImage image) Extract the PCA-projected VLAD feature from the given image.
byte[]	extractPQPcaVlad(List<? extends LocalFeature<?,?>> features) Extract the product-quantised PCA-projected VLAD feature from the given raw local features.
byte[]	extractPQPcaVlad(MBFImage image) Extract the product-quantisedPCA-projected VLAD feature from the given image.
LocalFeatureExtractor<LocalFeature<?,?>,MBFImage>	getExtractor()
FeatureVectorPCA	getPCA()
Function<List<? extends LocalFeature<?,?>>,List<FloatLocalFeatureAdaptor<?>>>	getPostProcess()
FloatProductQuantiser	getProductQuantiser() Get the product quantiser
VLAD<float[]>	getVLAD() Get the VLAD aggregator instance
int	index(List<? extends LocalFeature<?,?>> features, IncrementalFloatADCNearestNeighbours nn) Index the given features into the given nearest neighbours object by converting them to the PCA-VLAD representation and then product-quantising.
int	index(MBFImage image, IncrementalFloatADCNearestNeighbours nn) Index the given image into the given nearest neighbours object by extracting the PCA-VLAD representation and then product-quantising.
int	numDimensions() Get the dimensionality of the float vectors extracted from the pca-vlad process.
static VLADIndexerData	read(File file) Read a VLADIndexerData object to the given file created with the write(File) method.
static VLADIndexerData	read(InputStream is) Read a VLADIndexerData object to the given stream created with the write(OutputStream) method.
void	write(File file) Write this VLADIndexerData object to the given file.
void	write(OutputStream os) Write this VLADIndexerData object to the given stream.

Methods inherited from class java.lang.Object

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

Constructor Detail

VLADIndexerData

public VLADIndexerData(VLAD<float[]> vlad,
                       FeatureVectorPCA pca,
                       FloatProductQuantiser pq,
                       LocalFeatureExtractor<LocalFeature<?,?>,MBFImage> extractor,
                       Function<List<? extends LocalFeature<?,?>>,List<FloatLocalFeatureAdaptor<?>>> postProcess)

Construct with the given data

Parameters:

vlad - the VLAD extractor

pca - the PCA basis

pq - the product quantiser

extractor - the raw local feature extractor

postProcess - the process to apply to the raw features before VLAD aggregation

Method Detail

extractPcaVlad

public float[] extractPcaVlad(List<? extends LocalFeature<?,?>> features)

Extract the PCA-projected VLAD feature from the given raw local features. The local features will be post-processed before being aggregated using VLAD and projected by the PCA basis.

Parameters:

features - the raw local features

Returns:

the pca-vlad aggregated representation of the image

extractPcaVlad

public float[] extractPcaVlad(MBFImage image)

Extract the PCA-projected VLAD feature from the given image. The local features will be post-processed before being aggregated using VLAD and projected by the PCA basis.

Parameters:

image - the image to extract from

Returns:

the pca-vlad aggregated representation of the image

extractPQPcaVlad

public byte[] extractPQPcaVlad(List<? extends LocalFeature<?,?>> features)

Extract the product-quantised PCA-projected VLAD feature from the given raw local features. The local features will be post-processed before being aggregated using VLAD and projected by the PCA basis.

Parameters:

features - the raw local features

Returns:

the product-quantised pca-vlad aggregated representation of the image

extractPQPcaVlad

public byte[] extractPQPcaVlad(MBFImage image)

Extract the product-quantisedPCA-projected VLAD feature from the given image. The local features will be post-processed before being aggregated using VLAD and projected by the PCA basis.

Parameters:

image - the image to extract from

Returns:

the product-quantised pca-vlad aggregated representation of the image

getProductQuantiser

public FloatProductQuantiser getProductQuantiser()

Get the product quantiser

Returns:

get the product quantiser

createIncrementalIndex

public IncrementalFloatADCNearestNeighbours createIncrementalIndex()

Create an IncrementalFloatADCNearestNeighbours pre-prepared to index data

Returns:

a new IncrementalFloatADCNearestNeighbours

index

public int index(List<? extends LocalFeature<?,?>> features,
                 IncrementalFloatADCNearestNeighbours nn)

Index the given features into the given nearest neighbours object by converting them to the PCA-VLAD representation and then product-quantising.

Parameters:

features - the features to index

nn - the nearest neighbours object

Returns:

the index at which the features were added in the nearest neighbours object

index

public int index(MBFImage image,
                 IncrementalFloatADCNearestNeighbours nn)

Index the given image into the given nearest neighbours object by extracting the PCA-VLAD representation and then product-quantising.

Parameters:

image - the image to index

nn - the nearest neighbours object

Returns:

the index at which the features were added in the nearest neighbours object

write

public void write(File file)
           throws IOException

Write this VLADIndexerData object to the given file. The file can be re-read using the read(File) method.

Parameters:

file - the file to write to

Throws:

IOException - if an error occurs

write

public void write(OutputStream os)
           throws IOException

Write this VLADIndexerData object to the given stream. The read(InputStream) can read from a stream to reconstruct the VLADIndexerData.

Parameters:

os - the stream

Throws:

IOException - if an error occurs

read

public static VLADIndexerData read(File file)
                            throws IOException

Read a VLADIndexerData object to the given file created with the write(File) method.

Parameters:

file - the file to read from

Returns:

the newly read VLADIndexerData object.

Throws:

IOException - if an error occurs

read

public static VLADIndexerData read(InputStream is)
                            throws IOException

Read a VLADIndexerData object to the given stream created with the write(OutputStream) method.

Parameters:

is - the stream to read from

Returns:

the newly read VLADIndexerData object.

Throws:

IOException - if an error occurs

getVLAD

public VLAD<float[]> getVLAD()

Get the VLAD aggregator instance

Returns:

the VLAD aggregator

numDimensions

public int numDimensions()

Get the dimensionality of the float vectors extracted from the pca-vlad process.

Returns:

the dimensionality.

getPCA

public FeatureVectorPCA getPCA()

Returns:

the pca

getExtractor

public LocalFeatureExtractor<LocalFeature<?,?>,MBFImage> getExtractor()

Returns:

the extractor

getPostProcess

public Function<List<? extends LocalFeature<?,?>>,List<FloatLocalFeatureAdaptor<?>>> getPostProcess()

Returns:

the postProcess