[libav-bugs] [Bug 919] New: avconv does not match ffmpeg output for RGB24->YUVj444p conversion

bugzilla at libav.org bugzilla at libav.org
Fri Jan 22 07:23:25 CET 2016


            Bug ID: 919
           Summary: avconv does not match ffmpeg output for
                    RGB24->YUVj444p conversion
           Product: Libav
           Version: git HEAD
          Hardware: X86
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: ---
         Component: libswscale
          Assignee: bugzilla at libav.org
          Reporter: tgoodall at utexas.edu

Created attachment 573
  --> https://bugzilla.libav.org/attachment.cgi?id=573&action=edit
Everything needed to reproduce this problem, along with my output

Using the command line tool avconv, I want to convert some rgb24 data to a yuv
format like yuv420p or yuvj444p. Of course, I would like the conversion process
to be accurate.

Using a single frame of raw rgb video (from the bigbuckbunny test sequence), I
tested the rgb->yuv conversion process with ffmpeg. Now, given the converted
yuv file, I converted that back to RGB, then compared to the original RGB data
frame, to obtain a mean-squared-error value (less than 1).

The same test performed using libav's avconv obtained a higher
mean-squared-error value (nearly 2). This leads me to believe a small
fixed/floating point error exists in avconv.

Further analysis showed that the RGB24 -> YUV conversion operation does not
match bitwise, comparing libav and ffmpeg results. Also, it doesn't matter if
you change yuvj444p to yuv420p, ffmpeg and avconv still do not agree.

I tested this on the head and several previous versions of libav with the same
outcome. Also tried several versions (including the head) of ffmpeg. Please see
the attached script, raw data, and output from running said script.

You are receiving this mail because:
You are watching all bug changes.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.libav.org/pipermail/libav-bugs/attachments/20160122/090113ad/attachment.html>

More information about the libav-bugs mailing list