GNOME Bugzilla – Bug 161675
[ffmpegcolorspace] conversion of RGB depth=15 is wrong
Last modified: 2005-01-15 17:58:04 UTC
Conversion from video/x-raw-rgb, bpp=(int)16, endianness=(int)1234, depth=(int)15, red_mask=(int)31744, green_mask=(int)992, blue_mask=(int)31, width=(int)320, height=(int)240, framerate=(double)30, pixel-aspect-ratio=(fraction)1/1 to video/x-raw-rgb, width=(int)320, height=(int)240, framerate=(double)30, bpp=(int)32, depth=(int)24, red_mask=(int)65280, green_mask=(int)16711680, blue_mask=(int)-16777216, endianness=(int)4321, pixel-aspect-ratio=(fraction)1/1 gives incorrect colors. Tested with videotestsrc ! ffmpegcolorspace ! ximagesink.
http://bugzilla.gnome.org/show_bug.cgi?id=131974? I thought this was fixed now. I can, however, reproduce it. Given that the pipeline in 131974 works for me, is this a bug in videotestsrc?
Unless someone has since changed videotestsrc, it was tested with a 15-bit X server.
Horseb~1.avi on Stephane's FTP triggers this bug as well.
Created attachment 36058 [details] [review] Patch fixing the bug I'm quite certain this patch is right, but my understanding of colorspaces is pretty limited, so I may be wrong. However, it DOES fix the problem for me. For now, only the ->RGB15 case was done, the <-RGB15 case might need a similar change.
From the macro, it seems like this will break RGB16 code. Can you double check that and make a separate macro if not?
Created attachment 36062 [details] [review] Fixed patch The previous patch did break RGB16.. Further inspection showed that RGB565 was selected as the target colorspace instead of RGB555 (gst_ffmpeg_caps_to_pixfmt only checked bpp which is 16, not depth which is 15). Now, either bpp should be 15 (which is wrong, we use 16 bits to store it), or depth should be checked too. This patch adds code to gst_ffmpeg_caps_to_pixfmt does something like this: if (bpp == 16) { pixfmt = RGB565; if (depth == 15) pixfmt == RGB555; } That should not break RGB16, and works on RGB15 (will need to test RGB16 some way or the other.. any ideas?)
Oh, nicely noticed. Yes, this is indeed the problem. Applied, thanks.