Does Canon ever talk to their customers?

[Dermot Shane]

Alsitar;
you asked;
"Which is more "Raw" non de-bayered but compressed or de-bayered but uncompressed (F3)?"

having a rare chance to have an entire feature shot on each in my suite at the same time... definatly the flexibality award goes to the r3d's... the F3 is an entirely awesome camera in it's own right, but does not have the same level of malibality in post

I typicaly have 10 -14 days to grade a film, and it is really very common for the director, DP & myself to vary the look substaintly while we try things out... raw is a real world helper in this, as is having the raw in my timeline so i can re-set debayer choices interactivly before heading to fine tune in the software.

We are somewhat more limited in where we can go with the F3's footage... not impossiable, but certainly slower and certainly harder to play "what if"...

I don't have alot of experience with 8bit/422/compressed footage in this enviroment.. and what experience i do have has taught me to be very cautious, there is hard limits to run up against, and one does not have to go far to reach them.

Everything changes if you are finishing in 709 for broadcast & web tho... there i would expect the C300 to be in the same range as the XDcam's in terms of post flexibality... acceptable trade-off for the size and cost.

But the same trade-off's may not be worth the trade when working in P3.

Noise is not always the limiting factor, there are very good noise reduction options on the table now.

d

[Brian Drysdale]

I suspect most of the productions that the C300 will be used on won't have 10 to 14 days for the grading. Given how post production times are contracting and everything seems to become more squeezed, I gather 2 days for 45 mins screen time is a typical TV grading schedule.

[Jon Fairhurst]

Originally Posted by Alister Chapman

...IMHO the first thing that will make it hard to grade an image is noise....

Noise can actually make things easier. A good dither masks hard contours on smooth surfaces. The problem (as you noted later in your post) is too much compression. Compression can flatten the macro blocks. That actually hides the noise and creates gradients everywhere. In that situation, the noise no longer helps smooth the transitions. The noise also puts more stress on the codec so the quality goes down further. It can then add yet more errors to the flat macro blocks, causing them to dance between two bad levels. Uncompressed, it removes contours; compressed, it adds them.

But the noise isn't the problem. The problem is that compression is non linear and is less effective as noise rises.

[Peter Moretti]

Originally Posted by Alister Chapman

...
As for Red and "Raw" well if a lossy 5:1 compressed codec counts as "Raw" then so be it. It's only really "Raw" in so much as it has not been de-bayered. Which is more "Raw" non de-bayered but compressed or de-bayered but uncompressed (F3)?

Alister, I just want to point out that I had a similar discussion w/ a Sony engineer during an F65 event held at the DGA Theatre in Hollywood. Apparently the F65's RAW is lightly compressed (I imagine not as much as Red's is). The engineer mentioned that performing the compression on the mosaiced image yields better results than demosaicing first and then compressing. So at least it seems pretty clear that compressed RAW is better than compressed demosaiced.

As for uncompressed mosaiced vs. compressed RAW, I think it's important to look at the sensor pattern. If the underlying pattern is Bayer or some variant of Bayer, then uncompressed mosaiced has color interpolation and errors baked into it. Now this is okay if you are not going to heavy grade the image b/c the errors make sense w/ the composition.

But when you dramatically alter the look, the interpolation errors are now based on a look that no longer exists. Almost by definition, the errors will be in a direction that is not where you are moving the image to. (This is because they are based on the current look of the image and you are trying to alter that look significantly.)

So I believe to talk of a demosaiced 4:4:4 image as being uncompressed is kind of a misnomer. In fact, demosaicing itself is a form of lossy decompression.

[Peter Moretti]

Originally Posted by Dermot Shane

...
Most everything i have worked on has been shot on RED, film, or Alexa in the last few years...

Dermot,

May I ask, does Avid DS work natively w/ ArriRaw or are you using ProRes? I've heard that ArriRAW adds qutie a lot to the Alexa's image (if that even sounds possilbe).

[Dermot Shane]

Originally Posted by Peter Moretti

Dermot,

May I ask, does Avid DS work natively w/ ArriRaw or are you using ProRes? I've heard that ArriRAW adds qutie a lot to the Alexa's image (if that even sounds possilbe).

Peter;
Yes, DS does support Arri RAW, has since the D21 days, but i don't have any direct experience with Arri RAW, the shows i finished / graded were all ProRez targeted for 1080/709 deliverables... makes for a decent match.

This week is the first time i have worked with Alexa for DCi, and it was only a few pick-up shots, i got these in ProRez as well tho.

After a week with the F3/Slog/P3/DCi show we did a fairly detailed first pass in P3 with the DP, and then a 709 trim pass and sent it off for producer notes, some second unit was shot on Alexa, and i could not have told what camera it was but for the clip names in the timeline.. Arri's "LogC" and Sony's "Slog" are really close in the hands of a great crew.

The RED show got final tweaks and then the 709 trim pass, there is advantages to working off 4K files when targeting 2048x858, and then making a 16x9 1920x1080 show out of it, means working with one scaling operation only, and switching timeline settings to automagicly re-scale from the 4K. Simple and clean!

d

[Alister Chapman]

My comments on noise and grading were based on real world tests with codecs including 35/50/100 Mb/s Mpeg 2, ProRes HQ and DNxHD. In every case, the source noise was more important to the end result than the codec used until you got into noise figures better than around the 58db mark. With less noise than that then codec choice does become more critical.

When you have a low noise camera and you start to push and pull the image the noise almost inevitably becomes more visible. Yes, there are noise reduction software processes, but invariably these also reduce micro contrast and resolution, even the very best noise reduction programmes struggle to tell the difference between noise and things like rain, fluttering foliage and subtle moving textures.

Whether un-compressed de-mosaiced or compressed bayer data is better, this will depend on many factors and I'm no expert, but I suspect a lot will depend on the compression ratio, motion and the colour palette of the scene and quality of the de-bayer algorithm. The issue with de-bayering is that you have more data after de-bayer than before and perhaps this ends up as the key to why compressing pre de-bayer is preferable, especially at higher resolutions. Compression artefacts on the pre de-bayer luma signal will degrade the colour fidelity and depending on the artefact can affect an area greater than the original artefact due to the way the de-bayer process relies on adjacent pixels during it's calculations. Pro's and cons to both schemes as is often the case, these things are rarely black and white,

The key when your trying to extract the best from any camera is to understand to at least some degree what's going on under the hood. I'm surprised to hear that an S-Log F3 offers so much less grade-ability than a Red. I guess a like for like test is what's needed to really see how much difference there is. Different shots from different projects may not be telling the full story.

[Dermot Shane]

Originally Posted by Alister Chapman

I'm surprised to hear that an S-Log F3 offers so much less grade-ability than a Red. I guess a like for like test is what's needed to really see how much difference there is. Different shots from different projects may not be telling the full story.

I do think F3/Slog & AlexaLogC offer very good starting places for aggressive gradeing... and an brilliant starting place for more moderate gradeing as the skin tones look so very good straight away.. but i do think r3d's offer more flexiablilty...

I'm a huge fan of F3/Slog now that i have real world experience with it.

Originally Posted by Alister Chapman

even the very best noise reduction programmes struggle to tell the difference between noise and things like rain, fluttering foliage and subtle moving textures.

Agreed again.. just i'm not finding that noise is the limiting factor that it was a few years ago, it is better tamed with better tools now, and although i have not been feeling limited by noise in recent past sod's law will say the the next thing through the door will be a noisy disaster ;-)

Noise was a issue with a 3D project i graded with Si2K as the camera's a year ago... but that was the last time i have really had to make choices based on noise -vs- softness thankfully.

d

[Menno Mennes]

Dear readers,

In the Netherlands you can order the Canon C300 for:
Euro 12,000
GBP 8,269
US dollars 15,000
All prices are exc.VAT

nivo-schweitzer.nl

Have a nice weekend and best wishes,

Menno Mennes