I'm using QVideoFrame QVideoFilterRunnable::run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, RunFlags /*flags*/) However the video coming from an MP4 file is set as surfaceFormat.scanLineDirection() == BottomToTop. When I do analysis, I do it on a QImage. I have enough information that I can call mirrored() accordingly and generate an image that matches the frame. However if I alter that frame, and want to supply it down the filter line, I have to do my alterations, then re-flip then set it back as a QVideoFrame. This wastes a lot of time.
I feel a few things should be possible: 1. I can access QVideoFrame pixel data as if it were an image (QImage::pixel()) (Given some pixelFormat limitations, like ARGB and not classic (YUV) video formats) 2. I should be able to modify surfaceFormat and setScanLineDirection(TopToBottom) and feed in my TopToBottom image to QVideoFrame as no more unflipping required. Currently if I unflip then feed that, BottomToTop stays (surfaceFormat is const!) 3. Both QImage and QVideoFrame should be able to use a implicitly shared pixel data implementation, to make zero-copy a thing. 4. QPainter should be able to use QVideoFrames (again, given some pixelFormat limitations) Generally when I'm this off in the weeds, I'm doing something wrong. So is there a better way to do this? My frame processing is already taking 55ms (18fps), so adding unneeded flipping, back-flipping then who knows what the other steps of the pipeline take in terms of MS, which further reduce the framerate. Any suggestions? _______________________________________________ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
