image, based on the C# SSIM implementation by Chris Lomont. I see Avid now
offers some sort of PSNR plugin too, but I have no idea how it works (and I
haven't really tried).
On Wed, Jun 16, 2010 at 7:07 AM, nat jencks <natjencks.lists@gmail.com>wrote:
>
>
> Wow. Dylan, thanks for doing all the tests! What a public service!
>
> What tool did you use to measure the SSIM of the images?
>
> Again, thanks for posting your results, awesome work.
>
> best-
> -Nat
>
>
> On Jun 15, 2010, at 6:50 AM, Dylan Reeve wrote:
>
> > I got curious about this and went and tested the transcodes...
> >
> > Basically DNxHD 120, 185 and 185X were effectively identical to the
> source
> > clip. DNxHD 36 was very very good also, I couldn't see any differences
> > visually, and those I could detect were minimal. I also tested XDCAM EX
> > (35Mb/s) and XDCAM HD422 (50Mb/s) with strong results, again both
> appeared
> > visually identical in my tests.
> >
> > It seems that the detail retained in the H.264 is easily simple enough to
> be
> > effectively compressed into other codecs with minimal variance or loss,
> > especially where those codecs are of significantly higher baseline
> quality.
> >
> > I blogged it -
> >
> http://dylanreeve.com/videotv/avid/2010/transcoding-canon-dslr-footage.html
> >
> > On Tue, Jun 15, 2010 at 7:38 PM, Dylan Reeve <dylan@dylanreeve.com<dylan%40dylanreeve.com>>
> wrote:
> >
> >> In theory there should be no quality differences in decode methods. The
> >> only real differences should be in terms of decode time - I'm not sure
> how
> >> each of the applications do it, but I'd assume Adobe's apps are doing it
> a
> >> little differently as I believe their Mercury engine improves the decode
> >> times, using the power of the GPU.
> >>
> >> Canon's DSLR files confuse me, they are a pretty high bitrate (around
> >> 46Mb/s) but deliver pretty poor performance for that amount of data.
> They
> >> are 8bit and 4:2:0 sampled and tend to exhibit pretty noticable
> quantization
> >> artifacts. I'd tend to assume that any of the "full" DNxHD codec are
> more
> >> than capable of fully representing all the recorded image data (120, 185
> or
> >> 220).
> >>
> >> I'm fairly certain, however, that H.264 at such high bitrates should be
> >> able to do a lot better. If MPEG2 can deliver the quality it does with
> >> 50Mbit/s in XDCAM HD422, then H.264 at almost the same bitrate can
> surely be
> >> better?
> >>
> >>
> >> On Tue, Jun 15, 2010 at 4:13 PM, nat jencks <natjencks.lists@gmail.com<natjencks.lists%40gmail.com>
> >wrote:
> >>
> >>>
> >>>
> >>> In the course of researching various methods for processing 5Dmk2 and
> 7D
> >>> footage as dailies I have hit two questions for the lists collective
> wisdom.
> >>>
> >>> 1) Is there any difference between the various tools (Mpeg Streamclip,
> FCP
> >>> log and transfer, Magic Bullet Grinder, Adobe CS5, Neo Scene, etc) on
> the
> >>> DECODE side of things?
> >>>
> >>> From the research that I have done, all of these tools use the
> underlying
> >>> quicktime framework to decode the H.264, and any differences in
> transcoding
> >>> are the result of different types of encoding.
> >>> Possible exceptions are Neo Scene, and CS5, but even these may use the
> >>> same quicktime framework to decode?
> >>>
> >>> 2) On the ENCODE side of things, what is people's real world experience
> >>> regarding what type of datarate you need to go with to get ALL the
> >>> information out of these source files...
> >>> Prores HQ? Prores444? Uncompressed DPX? Obviously the safe thing to do
> is
> >>> go for an offline/online workflow with ProresLT for offline and
> Uncompressed
> >>> DPX for online, but if ProresHQ would capture all the source data
> without
> >>> introducing compression artifacts it would be nice to avoid
> offline/online
> >>> scenarios.
> >>>
> >>> Thanks in advance!
> >>>
> >>> Best-
> >>> -Nat
> >>>
> >>> [Non-text portions of this message have been removed]
> >>>
> >>>
> >>>
> >>
> >>
> >>
> >> --
>
> >> Dylan Reeve
> >> http://dylanreeve.com/
> >>
> >
> >
> >
> > --
> > Dylan Reeve
> > http://dylanreeve.com/
> >
> >
> > [Non-text portions of this message have been removed]
> >
> >
> >
> > ------------------------------------
>
> >
> > Please donate to the Red Cross to help those in earthquake ravaged Chile:
> http://tinyurl.com/y8pgceq
> >
> > Search the offical complete Avid-L archives at:
> http://archives.bengrosser.com/avid/
> >
> > Yahoo! Groups Links
> >
> >
> >
>
>
>
--
Dylan Reeve
http://dylanreeve.com/
[Non-text portions of this message have been removed]
------------------------------------
Please donate to the Red Cross to help those in earthquake ravaged Chile: http://tinyurl.com/y8pgceq
Search the offical complete Avid-L archives at: http://archives.bengrosser.com/avid/
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/Avid-L2/
<*> Your email settings:
Individual Email | Traditional
<*> To change settings online go to:
http://groups.yahoo.com/group/Avid-L2/join
(Yahoo! ID required)
<*> To change settings via email:
Avid-L2-digest@yahoogroups.com
Avid-L2-fullfeatured@yahoogroups.com
<*> To unsubscribe from this group, send an email to:
Avid-L2-unsubscribe@yahoogroups.com
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
No comments:
Post a Comment