darKoram's profile - activity

I found the test suite that I want at https://github.com/Itseez/opencv/tree/master/modules/java/android_test However, when i try to git clone it, i get fatal: https://github.com/Itseez/opencv/tree/master/modules/calib3d/tests/info/refs not found: did you run git update-server-info on the server?

I also tried git clone https://github.com/Itseez/opencv/tree/master/modules/calib3d/perf and got the same error.

truncate the above link to https://github.com/Itseez/opencv.git/ to be able to browse the repo and select what you want. Link above gives 404 onClick.

I'm trying to download the android test tools following http://code.opencv.org/projects/opencv/wiki/QA_in_OpenCV

When i try the following svn co http://code.opencv.org/svn/opencv/trunk/opencv_extra/testdata
svn co https://code.opencv.org/svn/opencv/trunk/opencv/moduies/java/android_test

I get svn: Could not open the requested SVN filesystem
and could not connect to server (https://code.opencv.org)

any tips?

According to the best practices page we want to use the fewest calls to the JNI as possible to get the job done.

In my application, I want to add a column of data (held in a List<double> or double[] to a Mat column. Rather than doing this

public void putColumn(int colIndex, List<Double> colData, MatOfDouble mat) {
    for( int i=0; i<colData.size(); i++) {
         mat.put(i, colIndex, colData[i]);
    }

I'd prefer to do something like this:

mat.putColumn(List<Double> colData);

Thus reducing the number of JNI calls from colData.size() to 1 call. Is there such a function in the Android API that does this, or do I have to build one with the JNI?

MatOfDouble has nice calls for exporting Mat data to java:
List<Double> mat.toList()
double[] mat.toArray()
but the fromArray(double...a) only seems to be for the constructor.

By the way, why the difference in declarations for float and double versions of put in MatOfDouble?

   public int put(int row, int col, double... data)

   public int put(int row, int col, float[] data)

I'm a c++ programmer and "double...a" is not something i'm familiar with. What does it mean?

It should be possible to create two submatrices within the parent matrix. submatrix Window0 could stay fixed at the start of the Parent matrix. Window1 could me the moving window. When Window1 bumps into the end of the Parent matrix, copy its contents into window0.

Window1.copyTo(Window0);

This should do the trick, since the dimensions and type match, create() should just copy headers.

"JNI calls are really expensive and should be minimzed" http://opencv.org/platforms/android/android-best-practices.html so that's extra incentive to find a call from the native library that works for my case.

Looking at the source code for Mat::copyTo in copy.cpp and Mat::create in matrix.cpp it appears that create() only allocates memory if the dimensions or type don't match. So trying to call M.copyTo(submatrixOfM) would cause a problem because create() would see dimension mismatch and try to allocate data within an already alocated space, or allocate the new data outside of the parent matrix (not sure which).

Hi, I'm working on an application that does some DSP for streaming data. OpenCV Mat seems like a pretty elegant solution for implementing a circular buffer inside a parent array. The moving window array will increment forward with each new dataRecord input as a column. When the front of the window (leading direction of motion) reaches the end of the parent matrix, I need to deep copy the window data to the start of the parent matrix. copyTo() looks good: http://docs.opencv.org/2.4/modules/core/doc/basic_structures.html#mat-copyto

The method copies the matrix data to another matrix. Before copying the data, the method invokes

m.create(this->size(), this->type);

so that the destination matrix is reallocated if needed. While m.copyTo(m); works flawlessly, the function does not handle the case of a partial overlap between the source and the destination matrices.

I'd like to use copyTo

public void wrapWindowToStart() {
            mWindowData.copyTo(mParentMatrix);
            //... extra code
        }

but based on the comments above i'm not sure if that will work. Does create() decide reallocation isn't needed and everything just works fine? For now, I'm doing something like this:

for (int i=mWindowBack; i<mWindowFront; i++) {
    for (int j=0; j<mRecordLength;j++) {
        mGpuDataBuffer.put(i-mQueueLength+mWindowLength, j, mGpuDataBuffer.get(i, j))
    }
}

OpenCV 2.4.2 worked great. The OpenCV Library Manager for android is fantastic. Linking libs is a real weak spot on android development and with the new manager everything was so smooth.

Fixes: 1. Eclipse seems to have an easy tool for selecting JRE within a workspace without changing system settings (via update-alternatives). Right-click project -> Build Path -> Configure Build Path -> click Libraries Tab -> Add Library -> JRE -> Search. You can then browse to the top level install location of your Sun JDK distro and Eclipse will find the files it needs. 2. The OpenCV-2.4.0 package shows no errors or warnings in Eclipse

Problems: 1. All sample projects except OpenCV-2.4.0 (library) have errors due to import org.opencv.<package>
This is despite explicitly including the library in the project.properties, AndroidManifest.xml and listing it as a referenced package. http://www.pasteall.org/pic/show.php?id=34473

The following shows some of the paths I've tried in my 15-puzzle/project.properties file

# Project target.
target=android-11

sdk.dir=/home/kesten/android-sdk-linux
# This project.properties file lives in OpenCV-2.4.0/samples/android/15-puzzle
# relative path to folder with OpenCV-2.4.0.jar file
# android.library.reference.1=../../../modules/java/android
# relative path to folder with OpenCV-2.4.0 project.properties file
# android.library.reference.1=../../../
#android.library.reference.1=/home/kesten/Android-OpenCV-2.4.1/OpenCV-2.4.0/modules/java/android/

I'm not sure if android can handle absolute paths to resources, but neither relative nor abs worked for me.

I was able to install the jdk from oracle.com .
http://www.oracle.com/technetwork/java/javase/install-linux-64-self-extracting-142068.html Select your flavour of distro. Mine: jdk-6u33-linux-x64.bin

It's possible that my problems installing before were from choosing an "RPM" type package. Instructions on another page indicate that Ubuntu prefers the auto-installer to rpms.

I followed the steps you gave. However, I have some concerns over using update-alternatives as it caused numerous problems in the past. (Changed phython to 3.2 for a project, but it broke nearly everything else in my distro).

It wasn't obvious in Eclipse where to locate the JRE. Selecting File->Build Path->Configure Build Path -> Select libraries tab -> Add Library -> System JRE Library -> Add [ I tried searching all through the /opt/java/64 hierarchy but Eclipse never recognized what it wanted. Finally, clicking on "Search" button instead of "Add" and Eclipse found it automatically.
http://www.pasteall.org/pic/show.php?id=34395

This didn't clear up my issues in Eclipse however. I tried File->Refresh and Project->Clean. Nope. I tried re-starting Eclipse... but it no longer starts:

[Popup dialog] An error has occurred. See the log file /home/kesten/.eclipse/org.eclipse.platform3.7.0155965261/configuration/1341530909313.log.

[log file] !ENTRY org.eclipse.osgi 4 0 2012-07-05 18:28:30.390 !MESSAGE Application error !STACK 1 java.lang.UnsatisfiedLinkError: Could not load SWT library. Reasons: no swt-gtk-3740 in java.library.path no swt-gtk in java.library.path Can't load library: /home/kesten/.swt/lib/linux/x8664/libswt-gtk-3740.so Can't load library: /home/kesten/.swt/lib/linux/x8664/libswt-gtk.so

at org.eclipse.swt.internal.Library.loadLibrary(Library.java:285)

libswt-gtk is installed, but it's in my java.library.path for openJDK, not the sun version.

I think I'd prefer a less "invasive" method that switches just the jdk/jre used for the project, rather than switching the whole system which may have unintended consequences. Will play with this idea.

BTW... I did get the commandline "$ ant debug" build working. Reading build.xml is very helpful. Was also able to fix the samples/android/fix_properties.sh script to work on my system. Will describe this in another post.