Changes in output when truncating bit-depth of RGB frames

[Tobias Rapp-2]

Hello,

when running some local regression tests for the upcoming v4.4 release I
noticed that the output checksum of some RGB test has changed. It seems
to happen when RGB input data having a bit-depth above 8-bit is
truncated to 8-bit and is caused by the change in commit
a48adcd1369131ebadd1c928fe790c72795182a7.

As the commit is targeted at RGB to YUV conversion I wonder why a
conversion like from gbrp12le to gbrp is affected by this change. Does
the swscaler temporarily convert to YUV when truncating RGB?

Example command:

ffmpeg -i FFVHuff_RGB12bit-720p25__Stereo-PCM24bit__FfMpegTestSrc.avi \
   -f framemd5 -map 0:v -filter:v "format=gbrp" -t 0.1 \
   -y /dev/stdout - report

Report file output is attached. The generated checksums are different
before/after the commit in git history.

Regards,
Tobias

_______________________________________________
ffmpeg-user mailing list
[hidden email]
https://ffmpeg.org/mailman/listinfo/ffmpeg-user

To unsubscribe, visit link above, or email
[hidden email] with subject "unsubscribe".