g3org3's profile - activity

@fab_cut Like, @berak said, please post a new-question, if you have further inquiries. Feel free to "at" mention me, if you would like my input.

@LBerger Obviously, I would like to use an existing one.

I have a function which takes in an cv::InputArray _inputArray, and I do a check for _inputArray.isVector(). Now, what I would like to do is directly use the cv::InputArray _inputArray as a vector, but it doesn't seem like there is overloaded functionality for this.

What is the correct way to go about doing this?

There is really no way to diagnose this problem without further information.

You can find them in the legacy support module: http://docs.opencv.org/3.1.0/d5/dc3/g....

My hope is that the cv::gpu::SolvePnPRansac modules will be moved to use cv::UMat, as that is more compatible and not CUDA specific. That is speculation, however.

Asaik, without re-writing some of the wrappers for the VideoCapture class, it does not work inside OpenCV4Android. I have tried several things without changing the source code. I think that it is because of the discrepancies and maintenance required to support various APIs/Devices. Also, maybe the underlying image buffers used by Android cause problems.

As the other answer indicates, you can use it to read from AVI, and some specified file formats. Apologies, I forgot to add that.

^ link is now broken, again.

I totally agree. The amount of help centered around building OpenCV4Android is really not good. But,it looks like it is trying to do some form of installation. But, in case you were wondering about usage, here is another question about the same thing, that was answered: http://answers.opencv.org/question/73...

Hope it helps!

Hey, have you ever solved this problem? I am having a similar issue with this. I am running a bootcamped Mac Book Pro, with Windows 10, and the run time performance of my application is significantly different than when run on my desktop version.

Actually seems like there is a fundamental problem with race conditioning. Like, I get consistent errors with vector iterators being out of bounds.... I feel as if there is maybe some bug with the OpenCL driver in boot-camped mac world.

As a temporary answer:

I have looked through the source code, and in cv::ocl:: there is specific statement that says OpenGL is only support for desktop platforms.

TL;DR:

What is the "official" way to build OpenCV4Android, from source, with OpenGL and OpenCL support?

I am using the Android NDK, and was using the pre-compiled OpenCV4Android binaries. Everything was going well, until I needed to do some texture handling, and had to use the cv::ogl::Buffer class. In this case, my code succeeded to build, because the cv::ogl header was referenced within the core module that is referenced in the CMake file within the pre-built binaries. However, I got the following error:

E/cv::error()(16616): OpenCV Error: No OpenGL support (The library is compiled without OpenGL support) in void {anonymous}::throw_no_ogl(), file /build/master_pack-android/opencv/modules/core/src/opengl.cpp, line 60

So, to get around this, I thought that I should re-build OpenCV4Android, trying to include the OpenGL flag, and (for posterity) explicitly turn on the OpenCL flag-- I have done custom builds on the desktop, no problem.

Building OpenCV4Android with OpenCL is actually pretty straight forward: after following this tutorial, you pass the -DWITH_OPENCL=ON to the .sh script for the target architecture type, and everything works fine.

However, building with OpenGL, seems really difficult. I know that there could be issues with the native OpenGL ES support that Android provides in their NDK, resulting in there being absolutely no-way to support OpenCV implementations of OpenGL. However, I find it weird that the header content for OpenGL would then still be referenced in the source code for Android builds. Also, after looking at the header file for the cv::ogl::Buffer class, it doesn't look like anything should conflict with the OpenGL ES content present in the Android NDK.

What I have tried so far is modifying the OpenGL flags within the parent CMakeLists.txt file in the main OpenCV directory, and the flag within the android.toolchain.cmake file; all with no success.

Any definitive answer on this would really be appreciated! Also, including this information within the development page, would save lots of people trouble! Knowing what they can, and cannot build with is important!

(Also, sorry for not hyperlinking keywords; apparently I have insufficient karma to do so. But, every link I followed was from the openCV website, under the Android Platform list. )