Re: Testing with lossless codecs
Re: Testing with lossless codecs
- Subject: Re: Testing with lossless codecs
- From: Brian Gardner <email@hidden>
- Date: Mon, 10 Aug 2009 12:44:21 -0700
Thanks, Darrin.
There appears to be no way to test my plugins using
a lossless QuickTime codec.
I tried the Difference matte technique you suggested to test a
number of QuickTime codecs, including the Animation codec.
None of them were lossless. Actually, it's kind of scary
that QuickTime does not have a single codec that is completely
lossless...
even the "uncompressed" were lossy.
I used a 10 bit DPX 1920x1080p 24.0 fps sequence as source footage,
which I've just received from a recent music video shoot, shot on a
stage.
The goal is to convert it into a single QuickTime ".mov" file,
with lossless compression (or even no compression). I couldn't achieve
that goal.
Here are the codecs that I tested:
Animation (at best setting) - truncates off the bottom 2 bits (10 bit
- > 8 bit), which is lossy.
None - Exact same results as Animation. 8-bit only. truncated bottom 2
bits. Lossy.
(Differencing the Animation and None codecs to each other gives 0's--
exact match.)
FCP Uncompressed 422 10 bit - differences at all edges (to be
expected, 444->422 issue)
DNxHD 10 bit 709 - Noisiest of these codecs. (Noisy into top 8 bits.)
Blocking artifacts,
especially in the vertically oriented direction.
Lossy.
Apple Pro Res HQ - slightly noisier than FCP Uncompressed 422, but
similar 444->422 lossy results.
Apple Pro Res 4444 codec - least lossy of the codecs I tested. Appears
to have saved full 10 bits.
(Did not just truncate down to 8. Possibly can do
16->16 bit?) Noisy in bottom bits, though. Lossy.
NOTE: The Apple Pro Res 4444 codec has a bug, in that it does not
appear to do
the right thing if it is given an image that does not have an alpha
channel.
And there is no option in the codec UI to automatically add a fully
opaque alpha channel.
(It would be very nice to have that option.)
When I layer two copies of the SAME resulting ProRes 4444 video on top
of each other,
and toggle visibility of the top layer On & Off -- they do not match!
There is a brightness/contrast difference, which appears to be
stemming from the alpha.
For my 10 bit 444 source video, Pro Res 4444 was less lossy than
"Uncompressed", "None", and "Animation".
"Lossless" QT codecs appear to only be lossless for 8 bit video, not
anything above 8 bits.
Ditto for 444 video ... where even "Uncompressed" is quite lossy. Both
were rather jarring surprises.
-- Brian
On Aug 10, 2009, at 10:08 AM, Darrin Cardani wrote:
On Aug 3, 2009, at 9:38 AM, Dan DiPaola wrote:
II have found that "NONE" does a fairly good job of holding a color
space. ... most of all no artifacts. If and when my gamma or
saturation shifts I do a compensation pass on the front end. I need
all the color gamut I can possibly get with ABSOLUTELY no
artifacts. We encode for Blu-ray, any artifacting will make itself
known on a 60' screen after we compress the HD down to 18Mbs. If
there is even so much as a hint of color banding or posterization I
will amplify it by a factor of two or more depending on the motion
and colors used in the shot. I do not believe Animation is a truly
and completely lossless codec.
The Animation codec absolutely is a truly and completely lossless
codec when set to Best temporal and spatial quality. If you find
that not to be the case, it's a bug. You can do a simple test by
compressing some footage with it, opening up the compressed footage
and placing it over the original footage and setting the blend mode
to Difference. If there are any pixels which are not 0, then it's
not lossless. (And if that's the case and you're sure you've set it
to Best quality, then it's a bug.)
Darrin
--
Darrin Cardani
email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Pro-apps-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Pro-apps-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden