Maximum Quality - Recommendations?

[Mike Finegan]

I am shooting on Scarlet and recently made the switch from Canon EF L series zooms to Primes (just Rokinon).
Did a lens test using the print it yourself chart at:
http://www.graphics.cornell.edu/~westin/misc/res-chart.html
Obviously the higher DPI the printer the more usable.

I discovered that at max spatial resolution - namely 4K WS 24fps - 5:1 compression, the image displayed on the monitor is distinctly sharper than what is actually recorded.
Granted 'visually lossless' depends on the content and distance, but it is big enough to cause me concern. Looks like 600 lines per screen height versus well over 1000.
No this is not scientific calibration - if that exists I am interested!

For Scarlet Dragon - it does not change much - according to the media guide 5K WS can only go down to 7:1, 4K WS is same as above 5:1. Seems only Epic Dragon gets big boost from minimag in that regard.

For Epic - 5K 24fps is 3:1
For Dragon - 6K WS 24fps is 4:1; 5K WS 24 fps is 3:1
Dragon with minimag - 6K 24fps - 4:1; 5K 24 fps is 3:1

So for my budget ($10k upgrade to Epic) seems that Epic is best choice for highest overall quality. Obviously less dynamic range than either Dragon, but extended range is mostly highlights.

Dragon, Epic Dragon, Weapon should all provide 'big screen' detail.
And before you say 'just shoot that scarlet' I am. Just have gotten some feedback about being soft from directors on occasion. Suspect that won't change much even using primes.
Just echoing some of the tradeoffs, from my perspective.

 

[Mike P.]

I discovered that at max spatial resolution - namely 4K WS 24fps - 5:1 compression, the image displayed on the monitor is distinctly sharper than what is actually recorded.
Granted 'visually lossless' depends on the content and distance, but it is big enough to cause me concern. Looks like 600 lines per screen height versus well over 1000.
No this is not scientific calibration - if that exists I am interested!

I'm confused by what you've found -- so 4k WS 5:1 (4096x1708) just **looks** like it is resolving almost double the amount of lines of 4k FF 6:1 (4096x2160) on the monitor/EVF, but the recorded image looks to be pretty much the same?

 

[Phil Holland]

The simple answer is:

Your lenses do make a difference. REDCODE RAW Compression Ratio makes a difference. Resolution makes a difference. The "thinner" OLPF on Dragon versus Mysterium-X makes a difference.

 

[Stacey Spears]

I would add to what Phil says, post sharpening. Probably something you want to do once you reach delivery size. The image in the LCD is being scaled down, which does sharpen the image. There is also a sharpness setting that makes the image in the display look sharper.

 

[Mike P.]

I would add to what Phil says, post sharpening. Probably something you want to do once you reach delivery size. The image in the LCD is being scaled down, which does sharpen the image. There is also a sharpness setting that makes the image in the display look sharper.

The sharpness setting is defaulted to 0. But, I still don't understand what he's asking -- I'm reading it as "4kWS 5:1 looks to have a higher spacial resolution than 4kFF 6:1 on the monitor/EVF, but the recorded image is pretty much the same"... Which seems rather pointless/meaningless, so I don't know why he'd be concerned.

 

[Marc Wielage]

I discovered that at max spatial resolution - namely 4K WS 24fps - 5:1 compression, the image displayed on the monitor is distinctly sharper than what is actually recorded.

What monitor? And how was it calibrated?

Note that apparent sharpness often hinges on image contrast, so judging the image before color-correction can be a very tricky subjective observation.

All the big lens-rental houses can get or loan you sharpness-testing charts that are a real torture test for all but the best lenses. I've seen the guys at Panavision calibrate their lenses, and it's a very tedious, time-consuming procedure, particularly for back focus and setting the focus ring on the lens.

Schneider Optics has some tips on how to use their focus chart here:

https://www.schneideroptics.com/pdfs/D101010.pdf

 

[Mike Finegan]

I was viewing on EVF Bomb but was using external monitor (1920x1080 60fps) to view detail. But could see similar detail in EVF also.
Not a calibrated monitor - but I do have a calibrated monitor I use for color grading - not sure that will change the B&W image. Did not see chromaic aberration - just 'focusing' on spatial resolution. Wish RED had 4K HDMI output then I could plug into 4K grading monitor or samsung suhd.
I used 4K WS because it allows the lowest compression ratio (for this camera - Scarlet X).
Just saying I could see much more detail on external monitor than in digitized stills.
BTW - with Panavision torture test mentioned - any hints on how many lines of resolution per picture height they can pull with better lenses on various RED cameras?
Looking to compare what I can do with best out there. Anyone with killer lenses and dragon care to provide a still of the chart link I provided? If printed at 600dpi should provide adequate target for this discussion. Phil - any possibility? You can PM is not appropriate to post...
Will look at the Schneider link.

 

[Mike P.]

Don't go by the monitoring path; go by the actual recorded .r3ds (which, as you said, the differences between 5:1 and 6:1 are negligible). Basically you're asking why your monitoring looks different between the two, but it's irrelevant since that's not what is being recorded.

Also, I'm pretty sure the live view is done at half-debayer, which will give perceived sharpness when downscaled to 1080 and/or viewed on a small screen like the EVF. There are too many variables here, and ultimately it doesn't amount to anything since the recorded files look the same/don't show a substantial difference.

 

[Jake Wilganowski]

What you are seeing is a 1/2 debayer on the monitor. Here's a little test to do which will enlighten you on red. Shoot something and take it into redcine. View it at 100 percent size, switch between full and 1/2 debayer. 1/2 will look sharper. Now switch to full and play with unsharp mask in the post effects tab.

RED IS NOT SOFT. It is just not over sharpened. They want you to apply however much sharpening you want. If your going for best quality, a full debayer will make a drastic difference in noise and color smoothness, then sharpen to taste.

 

[Phil Holland]

I was viewing on EVF Bomb but was using external monitor (1920x1080 60fps) to view detail. But could see similar detail in EVF also.
Not a calibrated monitor - but I do have a calibrated monitor I use for color grading - not sure that will change the B&W image. Did not see chromaic aberration - just 'focusing' on spatial resolution. Wish RED had 4K HDMI output then I could plug into 4K grading monitor or samsung suhd.
I used 4K WS because it allows the lowest compression ratio (for this camera - Scarlet X).
Just saying I could see much more detail on external monitor than in digitized stills.
BTW - with Panavision torture test mentioned - any hints on how many lines of resolution per picture height they can pull with better lenses on various RED cameras?
Looking to compare what I can do with best out there. Anyone with killer lenses and dragon care to provide a still of the chart link I provided? If printed at 600dpi should provide adequate target for this discussion. Phil - any possibility? You can PM is not appropriate to post...
Will look at the Schneider link.

If you want to see "cut your teeth sharp" detail, test at the appropriate distance on a specific focal length using either Zeiss Master Primes, Leica Summilux-C, and more recently the Zeiss Otus Primes. Shoot tests at wide open and two stops stopped down. Make sure critical focus is locked and very importantly before all of this make sure your back focus is calibrated accurately as that is important in getting the "maximum" possible performance.

Do this at the lowest compression ratio you can and perform the test at 4K, 5K, 6K and in between resolutions if desired. You will see (should see) visible detail gains shooting at 6K-5K and scaling down to 4K. Also, if you are getting dirty and deep into it, experiment with the various scaling options provided to you in REDCINE-X Pro. As Stacey recommended, experiment also with sharpening methods to see how much you can extract if you are really digging.

Somewhere in there you will find the answer you are looking for.

Right now based on the glass you mentioned you're shooting on I already know that's part of the limiting factor. I have the same glass here. And it's not just a matter of stopping that glass down. Glass quality and it's effect on light dispersion towards how it's "drawing" on the sensor is wildly different between budget minded and higher end lenses. You're looking at the difference between 140lp/mm to 200+lp/mm glass. Some of the top tier glass can resolve more than 200lp/mm.

If you have time and money one weekend, rent a well serviced set of Zeiss Master Primes or Ultra Primes and compare it to your Canon EF glass or Rockinons. One part of your answer is there and it's pretty clear. It's resolving power, contrast, and micro contrast you'll be seeing.

 

[Jake Wilganowski]

Micro contrast is huge and is one of the biggest differences between optics of poor and high quality

 

[Phil Holland]

Micro contrast is huge and is one of the biggest differences between optics of poor and high quality

it's one of the hardest things to "quantify". I remember once I was testing lenses prior to a large feature shoot with the DP and he described it nicely. "I'm not looking for too sharp, but I want it to grab the edges."

That stayed with me for a while.

 

[Steve Sherrick]

it's one of the hardest things to "quantify". I remember once I was testing lenses prior to a large feature shoot with the DP and he described it nicely. "I'm not looking for too sharp, but I want it to grab the edges."

That stayed with me for a while.

Indeed, this is one of the things that I notice with my set of glass. They can produce nice, pleasing images but sometimes I just need a bit more out of them. What that DP said to you is exactly what I feel sometimes.

 

[Mike P.]

I play with offset/gamma correction to exaggerate "edge grabbing"... Sometimes it works really well, but ideally the glass (and camera) would do it inherently. Actually, I've noticed that bigger pixels do it better, in my opinion.

But, yes, back to the topic; I think you're merely seeing something that doesn't exist and is a combo of live-view and monitor/evf.

 

[Phil Holland]

I play with offset/gamma correction to exaggerate "edge grabbing"... Sometimes it works really well, but ideally the glass (and camera) would do it inherently. Actually, I've noticed that bigger pixels do it better, in my opinion.

But, yes, back to the topic; I think you're merely seeing something that doesn't exist and is a combo of live-view and monitor/evf.

One interesting trend in still and motion picture processing has been the addition of the local contrast adjustment. For those familiar with Adobe Camera Raw, that is Clarity. It's also a common "enhancer" in DaVinci Resolve and other apps.

It's something you can certainly enhance in post, but there is a subtle effect of having it in the glass that's truly hard to nail outside of it being in the footage itself.

I'll tell you man. I've stared deeply into the heart of glass from both creative and technical perspectives and it's still one of those things you just end up getting deeper and deeper into.

This is coming from a guy who enjoys shooting on funky old stuff and ruthlessly clean modern stuff. Never one answer for me it seems.

 

[Anton Shavlik]

I remember, in a thread from long ago, Graeme said that full debayer has mathematically some different characteristics than the fractional debayer options, added sharpness being the major difference.

edit:

I'm not sure which of these applies to what is sent to the screen, first link implies that the sharper appearance in "half res good" is do to aliasing on the edges, second states "Half High (due to the sample method)... has some mild sharpening in it"

http://www.reduser.net/forum/showthread.php?55091-Half-Res-Good-vs-Full-Res-Premium

http://www.reduser.net/forum/showthread.php?46415-Odd-difference-between-Full-amp-Half-Res-high

In either case, the full debayer will have more detail after you add some sharpening / resizing

 

[Mike Finegan]

Thanks everyone for responding. Lots of experience being shared.

I will do the half bayer apparent sharper test mentioned, but I don't think that is it as there is no way to take a gray value and magically get uniform width bars out of it.
I flip back and forth between different debayer setting all the time when editing. If it was just edginess I would say that is probably it. But it is symmetric texture.
Anyway I will repeat the test with photos of the monitor and an r3d.

But in general I get that the monitor is not seeing the same debayer and sharpening level as what is captured in the r3d.
And definitely shoot at 5K or 6K and down res in post to 4K to get maximum information content (and apparent sharpness). So the 1080p is seeing that effect, but still 2:1 more information is a lot. Any sharpening tricks should not be better than a 2 times oversampling. At best equivalent. But maybe very aggressive sharpening is applied in camera for preview.

For contrast, the Rokinon primes appear more three dimensional than the Canon L zooms - and I believe that is due to better contrast / local contrast. In one of the red groups it was stated that that effect is even stronger on certain Zeiss primes, etc. Feel free to send a good link on that Part of the character of individual cine lenses.
Just was surprised by what I ended up with at 4:1 compression.

As suggested, I should rent red dragon and real cine lenses if I truly want to see the extremes. I have one day left to shoot for a project where I might use dragon and can do that test after main shooting. And I have not back focused this camera - always possible it is off.

 

[Mike Finegan]

Once again thanks for all the input.
Did a quick test using the target I sent the link for (it has to be printed at > 1200 dpi to avoid moire related blur).
The same lens (Canon L series zoom or equivalent Rokinon prime) is much sharper on 5DMk3 than Scarlet at 5K.
5DMk3 seems to go all the way up to the limit of the target (2000 lines per picture height) whereas for Scarlet at 5K it seems to top out around 1500.
Doesn't seem to be compression related as I tried every compression possible from 3:1 up tp 18:1 and there is not a huge difference. So some of the 'blur' is before the sensor if compression does not change dramatically (I am thinking).
Canon 5DMk3 is know for strong OLPF - is Scarlet OLPF even stronger? Or do I have back focus issue?
For anyone who does lens validation at rental house and shoots with Mysterium - what is the max lines per picture height you have seen? Scarlet 5K and Epic 5K (stills) are identical, right?

But to make a long story short the glass is not the limiting factor. So maybe I need to move to Scarlet Dragon, at least. Was hoping Epic at 5K would be at least as good as this glass.
Thx
Mike

 

[Stacey Spears]

OLPF on non-Dragon EPIC/SCARLET is very strong. Blurs lots of detail. I believe the Canon also has sharpening applied unlike the RED footage where you are expected to sharpen to taste.

Take a look at the two images in this zip file. http://www.spearsandmunsil.com/sspears70/dcon_04.zip one is the original, Dragon at 6K HD, and the other has had deconvolution applied to try and undo the blur caused by the OLPF. View at 1:1 and look at the signs on the side of the trash can, for example. The 3rd file is difference between the two so you can see the pixels that have been altered.

 

[Blair S. Paulsen]

Whatever "reference" monitor you might be looking at, it does have it's own particular dynamics. Capture resolution>math>actual raster of the display device means scaling operation(s) must be used which will alter edge definition, etc... Keep in mind there are lots of reasons that perceived sharpness in a particular signal>display>viewing scenario might not be a definitive representation. Be aware: hardware scaling may not be an exact match to software scaling so compare the two if you're starting anything long form.

What Stacey is doing is brilliant and I've often wondered how much of this type of processing is built into the debayering algorithm natively.

Cheers - #19