[mythtv-users] H264 conversion of interlaced MPEG2?

Dave MythTV dave.mythtv at gmail.com
Wed May 6 02:03:40 UTC 2015


Thanks everyone, a lot of great responses in this discussion, and I
certainly appreciate the help!


Mike Stucky:
I ran some tests with a modified version of your update, omitting the frame
rate specifier and using yadif=1:-1:1.   This worked great on interlaced SD
content... but it failed on my progressive HD sample.  I gave it a
progressive-scan 1280x720 59.940 fps input, and got a progressive scan
1280x720 output at 119.880 fps!    So that third yadif flag, "deint", which
I set to 1 meaning "Only deinterlace frames marked as interlaced" did not
do as I had expected.   'mediainfo' correctly identifies the source video
as progressive... so I'm not quite sure why that didn't work?  (In other
words, what does "frames marked as interlaced" actually mean for MPEG2
content within ffmpeg?)


Dennis:
Thanks for the seektable / seeking explanations!
I'll add -g 120 to my parameter list to give more frequent but a reasonable
number of keyframes, and try your suggestion for the scenecut threshold.
The '-movflags faststart' part is interesting, and definitely sounds
beneficial for my usage, so I'll add that one too.   The documentation says
it can take a while... have you seen a significant impact to encoding speed
from that option?
Thanks!


Mike Perkins:
Good information.  Hopefully things should be at least a *little* easier
for the players to sort out, as I should ultimately end up not needing any
deinterlacing at playback.


Michael Wisniewski:
Thanks for the script.  I'm certainly not tied to ffmpeg, so if something
else works better that would be fine with me.  Not being familiar with
HandBrakeCLI, it doesn't look like your script does anything specific to
manage the deinterlacing.  Does HandBrake do this automatically?  (In
particular, does HandBrake do it the "nice" way automatically, instead of
ffmpeg's default of just copying fields with artifacts into a progressive
container?)


f-myth-users:
Back when I was running an interlaced system end-to-end, I had a huge
problem with everything automatically cropping, shifting, and scaling for
overscan reasons.  The end result was that even on an output configured to
*exactly* the same resolution and refresh rate as the input, I didn't have
1:1 alignment of the scanlines, so everything was blurred from the
interpolation and looked bad... and like you said, with visible interlacing
artifacts even with interlaced content on an interlaced display.  At the
time with the hardware I had, some of these settings couldn't be changed
(or couldn't be changed enough).  We upgraded from a CRT TV to a large
workstation-grade CRT monitor (high resolution, progressive scan) and it
was amazing how much better the exact same recordings looked... even
without any smart deinterlacing for playback.

I later ran an interlaced setup driving a HD CRT projector directly with an
analog RGB connection.  For that, I had much better control... but by then
I was into the situation where I had a mix of interlaced SD content,
interlaced and progressive DVDs, and HD content at various resolutions.
For picture quality, I had the best luck when I deinterlaced and pre-scaled
ALL of the content after recording, but then let the display output
"re-interlace".   It worked really well, but that was a very specific case.

While you still have your original files, have you tried letting your
scripts fully deinterlace them (not just copy fields like your and my first
rounds did), and then playing those progressive files back on your
interlaced output?   Depending on your output configuration, the extra
lines/fields inserted from the deinterlace may just be discarded with very
little quality difference from the original...  but if there is any scaling
in the chain, like the overscan provisions, it might solve some of your
artifacts when it re-interlaces.

Good luck!



Thanks again everyone!
- Dave
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mythtv.org/pipermail/mythtv-users/attachments/20150505/12171392/attachment.html>


More information about the mythtv-users mailing list