Geek Answers Handbook

Fast conversion of YUV420P to RGBA on iOS using a processor

Can I recommend a really fast API, ideally NEON-optimized for converting YUV to RGB at run time on iPhone using a processor? The vImage acceleration view does not provide anything suitable, unfortunately, for using vDSP, converting to and from floats seems to be suboptimal and almost the same as the NEON record itself.

I know how to use the GPU for this with a shader, and actually already do this to display my main video plane. Unfortunately, I also need to create and save the RGBA textures of the display subregions at runtime. Most of the good answers to this question are related to shaders, but I don't want to use the GPU for this extra work, because:

(1) Although I could use RenderTextures and my YUV shader to convert and cache regions, I don’t want to add more synchronization / complexity to the application. (I already transfer textures from CVTextureCache to Unity3D ... I already in many cases rearrange the state from OpenGL to Unity3D and don’t want to debug anymore ...)

(2) I’m practically writing a game and I don’t have a GPU (as usual, there are no games). I gave some presentations on how to remove things from the GPU over the past few years than how to put things on it ...)

(3) On the iPad, I have a spare core, doing nothing.

As long as there are many libraries that will do YUV in RGBA, I would love to save the time writing my own version of NEON. Now I am using the OpenCV implementation as follows:

cv::cvtColor(avFoundationYUVCaptureMat, BGRAInputImage, CV_YUV420sp2BGRA, 4);

which is true, but at the end of the day slower.

If someone previously looked at other implementations (CoreImage? FFMpeg?) And can recommend one, I would be very grateful.

Thanks, Alex.

+7
source share
2 answers

If you have Y , U , V in separate planes and there is no downstream chromatization discretization, then you should use vImageMatrixMultiply_Planar8 to convert.

+4
source

All I did was google yuv rgb neon and that is what I found - https://code.ros.org/trac/opencv/browser/trunk/opencv/android/android-jni/jni/yuv2rgb_neon. c? rev = 4083

As the comments say, this is still rude, and you might have to adapt it to convert to RGBA, but that would not be easy.

As far as I know, other such features should be published on NEON blogs - checking hilbert-space, webshaker, a blog for turning on hand software, etc.

+2
source

More articles:


All Articles