wilsonharron / lire Goto Github PK

FuzzyColorHistogram#getDoubleHistogram()
throws an UnsupportedOperationException.

I'd expect to return the int array.
The patch includes an additional toDoubleArray method in the serialization 
class + an implementation of getDoubleHistogram()

https://code.google.com/p/lire/source/browse/trunk/src/main/java/net/semanticmet
adata/lire/utils/SerializationUtils.java#245


In the 0.9.4-beta2 code, the function seems to shift the output array by 
offset, and chop off the rest of the float array at the end.

    public static float[] toFloatArray(byte[] in, int offset, int length) {
        float[] result = new float[length / 4];
        byte[] tmp = new byte[4];
        for (int i = offset; i < length / 4; i++) {
            System.arraycopy(in, (i - offset) * 4 + offset, tmp, 0, 4);
            result[i] = toFloat(tmp);
        }
        return result;
    }

For comparison, the code for converting to double array is correct:
    public static double[] toDoubleArray(byte[] data, int offset, int length) {
        double[] result = new double[length / 8];
        byte[] tmp = new byte[8];
        for (int i = 0; i < result.length; i++) {
            System.arraycopy(data, i * 8 + offset, tmp, 0, 8);
            result[i] = toDouble(tmp);
        }
        return result;
    }

hi,I find that this project support lucene4.0 and the svn commit logs says 
that: sift test successfully.
so,I have a try.

I copy the src/test/java/...../lire/indexing/IndexVisualWordsTest.java  and run 
it.
but i found when the number of  images which will be indexed  are very small 
(for example , only 5 images should be indexed), then when run the code:
 -----------------
System.out.println("-< Creating SURF based histograms >--------------");
SurfFeatureHistogramBuilder surfFeatureHistogramBuilder = new 
SurfFeatureHistogramBuilder(IndexReader.open(FSDirectory.open(new 
File(index))), numSamples, clusters);
surfFeatureHistogramBuilder.index();  
---------------------
 there will be a bug:  when run k-means algorithm.
the console print:  k.getFeatureCount() = 0
 then there will be a NullPoint at:KMeans.init():
------------------------------
Set<Integer> medians = selectInitialMedians(numClusters);
Iterator<Integer> mediansIterator = medians.iterator();
------------------------------
if I debug this code, I can find that medians=null.


But when I add images to be indexed, for example 20 images, the issue will 
disappear.

Does this means surf need many images ?  and sift need more?(even though I use 
20 images to test sift algorithm, the NullPoint exception appears . )  

This patch fixes the compilation errors in 
FloatArray2DScaleOctaveDoGDetector.java by using Commons Math 3 library for 
ArrayMatrix handling, rather than relying on version 2 of library.

What steps will reproduce the problem?

CEDD cedd = new CEDD();
cedd.Compact = true;
cedd.extract( ImageIO.read(new File("c://myfile.jpg")));
System.out.println(cedd.getStringRepresentation()); 

What is the expected output? What do you see instead?

desired:
cedd 60 5 3 1 3 1 0 0 0 0 0 1 2 0 0 2 0 0 0 0 0 4 6 0 3 4 0 0 0 3 0 1 3 0 1 4 0 
0 0 1 0 0 1 0 0 1 0 0 0 0 0 1 2 0 0 1 0 0 0 0 0

instead: (filling with 0s)
cedd 144 5 3 1 3 1 0 0 0 0 0 1 2 0 0 2 0 0 0 0 0 4 6 0 3 4 0 0 0 3 0 1 3 0 1 4 
0 0 0 1 0 0 1 0 0 1 0 0 0 0 0 1 2 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0


What version of the product are you using? On what operating system?
0.9.4 (trunk)

Please provide any additional information below.

If thats how it should be the tanimoto similarity is slower than having only to 
loop 60 bits

in function createShape()
Line 193: int[] pixel = {0, 0, 0};

If the input image with alpha channel
Exception in thread "main" java.lang.ArrayIndexOutOfBoundsException: 3
    at java.awt.image.ComponentSampleModel.getPixel(Unknown Source)
    at java.awt.image.Raster.getPixel(Unknown Source)
    at net.semanticmetadata.lire.imageanalysis.mpeg7.ColorLayoutImpl.createShape(ColorLayoutImpl.java:196)

Suggestion:
int[] pixel = {0, 0, 0, 0};

What steps will reproduce the problem?
1. add only one image to index with help of CEDD documentbuilder
2. search by this image (note, same image as you added to index)
3. in ImageSearchHits you will have only one record, calling hits.score(0) will 
throw java.lang.AssertionError exception

What is the expected output? What do you see instead?
expected - 0; exception instead

What version of the product are you using? On what operating system?
0.9.3, Windows 7x64

Please provide any additional information below.
I attached source files for your convenience 

thanks

Hi all

Maven's <em>interesting</em> xml parser seems to dislike the lire copyright 
header, attached is a very small patch that hopefully makes this work a little 
better.

Hi, I am using lire.jar as a library in one of my applications. I need to 
serialise some classes that contain lire classes implementing the LireFeature 
interface, but currently that is not possible because the lire classes are not 
serialisable. Could the LireFeature interface be modified to implement 
serialisable? Alternatively the individual classes implementing LireFeature 
could implement serialisable instead.

Thanks
Mike

in 
net.semanticmetadata.lire.imageanalysis.sift.Feature.getByteArrayRepresentation 
method
descriptor is a double array. so that SerializationUtils.toBytes(descriptor[0]) 
return 8 bit byte array. but only first 4 byte is using to serilazing and 
deserilazing.

i attached patch code.

1. StringField is not tokenized, no matter stored or not(lucene 4.2.1)
2. line 48 in FieldType.java :private boolean tokenized = true;
3. so, any StringField will be tokenized if display it's filedtype. would some 
mistakes occurs？
4. until lucene 4.8.0 ，keep unchanged .

Add generics type-safety to code.

What steps will reproduce the problem?
1. Run the Lire-demo
2. Index a database
3. Search with SURF BOvW

What is the expected output? What do you see instead?
No results, even though the features were extracted and saved. 

What version of the product are you using? On what operating system?
LireDemo-0.9.4-beta_2.zip on Ubuntu 12.04

Please provide any additional information below.
I tried to work out examples with Sift/Surf/MSER using VisualWordsImageSearcher 
class. But most of the extraction methods are not implemented. I guess the 
search mechanism is different for bang of words kind of features. Can someone 
let me know the status of the aforementioned features and if they are in place, 
how can I make use of them.

regards
Mohit

This patch includes the fixes to the following issues:

* fixes warnings on used code
* fixes dead code warning
* fixes an obvious null pointer exception in GenericDocumentBuilder
* adds type-safety by removing raw-type usage
* fixed resource leak by closing all streams
* fixed lucene IndexReader.open() deprecation warnings

Hope this helps.

int line 87:
System.arraycopy(descriptor, 0, clusters[i].mean, 0, descriptor.length);

clusters[i].mean is float[], but descriptor is double[], so it will cause 
java.lang.ArrayStoreException

I am use version 0.9.3

This patch upgrades the Maven POM file for the following:

1. Remove a parsing error by removing '--' from the copyright header and 
replace it with '=='
2. Use a Mavenized and upgraded 'jopensurf' library so as not to rely on system 
dependency
3. Add the WEKA library dependency for preventing missing dependencies

The source codes of net.semanticmetadata.lire.clustering.Image.java and 
net.semanticmetadata.lire.imageanalysis.sift.Found.java are missing in the zip 
file

The line 152 of Class LocalFeatureHistogramBuilder.java is
if (features.size() < numClusters)
but I think it should be 
if (k.getFeatureCount() < numClusters)
because features only stores one images features.

net.semanticmetadata.lire.imageanalysis.LocalBinaryPatterns cannot be used (it 
looks like it has been deactivated)

What steps will reproduce the problem?
1. add lire.jar v0.9.5 to a project
2. use the ParallelExtractor to extract the Local Binary Patterns feature from 
a set of images

What is the expected output? What do you see instead?
The extraction process raises a NullPointerException

What version of the product are you using? On what operating system?
Code running under Intellij Idea 13.0.1 on Windows 7 SP1 64 bit

Please provide any additional information below.
The problem lies in net.semanticmetadata.lire.indexing.tools.ParallelExtractor 
where the LocalBinaryPatterns is not added in the member called "features":

    public static final String[] features = new String[]{
            "net.semanticmetadata.lire.imageanalysis.CEDD",                  // 0
            "net.semanticmetadata.lire.imageanalysis.FCTH",                  // 1
            "net.semanticmetadata.lire.imageanalysis.OpponentHistogram",     // 2
            "net.semanticmetadata.lire.imageanalysis.joint.JointHistogram",        // 3
            "net.semanticmetadata.lire.imageanalysis.AutoColorCorrelogram",  // 4
            "net.semanticmetadata.lire.imageanalysis.ColorLayout",           // 5
            "net.semanticmetadata.lire.imageanalysis.EdgeHistogram",         // 6
            "net.semanticmetadata.lire.imageanalysis.Gabor",                 // 7
            "net.semanticmetadata.lire.imageanalysis.JCD",                   // 8
            "net.semanticmetadata.lire.imageanalysis.JpegCoefficientHistogram",
            "net.semanticmetadata.lire.imageanalysis.ScalableColor",         // 10
            "net.semanticmetadata.lire.imageanalysis.SimpleColorHistogram",  // 11
            "net.semanticmetadata.lire.imageanalysis.Tamura",                // 12
            "net.semanticmetadata.lire.imageanalysis.LuminanceLayout",       // 13
            "net.semanticmetadata.lire.imageanalysis.PHOG",                   // 14
    };

I could fix by just adding 
"net.semanticmetadata.lire.imageanalysis.LocalBinaryPatterns" as additional 
entry.