5k or 2.5k...?

[Álex Montoya]

While shooting with a stock Red can get about 3.5K, I go well beyond that.

Here is a shot that was shot 4K , Canon 17-55 f2.8 at F5.1. Handheld shot.
Deleted OLPF and special post processing is my modus operandi .

Go 6K

link to 6K jpeg, please load into capable image viewer to zoom 200%.

http://www.mediafire.com/view/?8hfytb52qjjgofb

Wow, that's plain amazing.

 

[Les Dittert]

Wow, that's plain amazing.

Thanks. But most users don't care for high detail right now. They want creamy-organic-insert-term looks provided by old glass and such. Works OK for 1080.
The widow maker will be the UHD TV sets where even non technical types will see what's up.
Resolution isn't everything, obviously, but after seeing a genuine 4K film ( Samsara ) projected, I'm an addict of details.

 

[David Mullen ASC]

The 4K screening of the last ASC-PGA workflow assessment series basically showed what you would expect (based on the sensor specs and finishing at 4K), more or less, in terms of visible detail, it was something like: 35mm tied with Alexa, then Red Epic, and finally Sony F65 in terms of increasing levels of fine detail, but I would also say that they were all very close in terms of final image quality.

 

[Sidney L. Plaut]

David,

just to be sure I understand - which had the highest level of detail - the F65 or Alexa/35mm?

 

[Noah Yuan-Vogel]

Great to know, thanks. However that bolded part really scares me... is it really performing that much lower? I mean that essentially means it's worse than dslr footage since it would have the same range and resolution but with a lot more noise...

I'm not sure I know exactly what you mean specifically, but wouldn't it be? It may not be so simple, since even if it mathematically makes sense. If pixels are blown up 200%, is noise increased by 200%? It depends how you measure it and how you judge it. What do you mean worse than DSLR footage? Dynamic range is essentially defined by noise, especially a RED camera with such customizable gamma curves and RAW processing. Same (dynamic) range and resolution with more noise? How do you mean? if it has more noise, doesn't it have less dynamic range? Just speaking very roughly, if RED 4k has 13.5stops of DR and RED 2k has 11.5 stops, I don't know of many DSLRs that could be said to have more than 11stops of DR when shooting video... So you may still come out ahead with RED 2k, but that may change a bit if you have to shoot higher than ISO400.

 

[James_Mills]

I'm not sure I know exactly what you mean specifically, but wouldn't it be? It may not be so simple, since even if it mathematically makes sense. If pixels are blown up 200%, is noise increased by 200%? It depends how you measure it and how you judge it. What do you mean worse than DSLR footage? Dynamic range is essentially defined by noise, especially a RED camera with such customizable gamma curves and RAW processing. Same (dynamic) range and resolution with more noise? How do you mean? if it has more noise, doesn't it have less dynamic range? Just speaking very roughly, if RED 4k has 13.5stops of DR and RED 2k has 11.5 stops, I don't know of many DSLRs that could be said to have more than 11stops of DR when shooting video... So you may still come out ahead with RED 2k, but that may change a bit if you have to shoot higher than ISO400.

Not sure what you mean by stating that dynamic range is measured by noise; how does noise dictate how many shades of grey are apparent?

Anyways, I hate to digress but this leads me to another question in regards to the changing resolution rates on RED cameras: since the EPIC and Scarlet are running off the same sensors, is the difference in noise / iso / range that we're talking about from 5k (epic) to Scarlet (4k) the same as a scarlet shooting at 3k instead of 4k? For instance, as I mentioned earlier, when you shoot at 3k or 2k on a Scarlet you're getting more noise and the image doesn't look as good (not 3k as much, but still somewhat). Is this the same difference from 5k on an Epic to 4k on a Scarlet? If so, why do they both take up the same amount of sensor space (both are still shooting at a 35mm full frame size, right? Or is 5k full frame a bigger frame than the Scarlet's 4k full frame?)

I've been wondering this because so many people say that theoretically the Epic's image and the Scarlet's should be identical since they're both the same sensor, but this shouldn't be the case if we're seeing similar dropoffs across the board as we do on a Scarlet when shooting at higher framerates...

 

[Björn Benckert]

It's quite simple epic shoot Cmos 5k 16bit compressed raw. that resolves to about 12bit 4k when processed.
Alexa without raw recorder records Cmos 2,8k 10bit proress. That image when processed resolves to HD 8 bit.

So epic has more colors and higher resolution not only in pixel count but also in actual / measured resolution. So basically your VFX friend have not done his home work correctly.

However the alexa does a double readout as it's standard readout from the sensor and by doing so it gets almost 2 stops more range than the epic when shooting without HDRx. However the epic with HDRx on can possibly yield an even higher DR than alexa but the epic can then start to show artifacts when the HDRx is turned up to high.

But all in all I would say that the technical spec and tests of the epic beats the alexa even though the alexa has a very high "default level". Simply put, it's way easier to shoot shit with the epic, but right hands the epic stands way above the alexa. The F65 is a completely different beast... We need the dragon before we want to start to comparing the epic with that one... lol.

 

[Noah Yuan-Vogel]

Not sure what you mean by stating that dynamic range is measured by noise; how does noise dictate how many shades of grey are apparent?

Anyways, I hate to digress but this leads me to another question in regards to the changing resolution rates on RED cameras: since the EPIC and Scarlet are running off the same sensors, is the difference in noise / iso / range that we're talking about from 5k (epic) to Scarlet (4k) the same as a scarlet shooting at 3k instead of 4k? For instance, as I mentioned earlier, when you shoot at 3k or 2k on a Scarlet you're getting more noise and the image doesn't look as good (not 3k as much, but still somewhat). Is this the same difference from 5k on an Epic to 4k on a Scarlet? If so, why do they both take up the same amount of sensor space (both are still shooting at a 35mm full frame size, right? Or is 5k full frame a bigger frame than the Scarlet's 4k full frame?)

I've been wondering this because so many people say that theoretically the Epic's image and the Scarlet's should be identical since they're both the same sensor, but this shouldn't be the case if we're seeing similar dropoffs across the board as we do on a Scarlet when shooting at higher framerates...

Dynamic range in a properly designed digital camera system is generally the range that exists between the white clip and noise floor of the system. If you crop into a 2k portion of the RED MX sensor, you are magnifying the noise, making it more visible. It would stand to reason that if you are using a portion of the sensor only 1/4 the size of the 4k portion it should have 4x more noise, which translates to 4x higher noise floor, which means two stops less sensitivity and two stops lower you have to rate the camera while keeping the same white clip point as you would have at 4k.

Yes 5k is a larger sensor area than 4k, but it is 1/4 more sensor area, whereas 4k is only 1/3 more than 3k and 3k is 1/2 more than 2k, so there is a similar effect with all of these steps but to differing degrees.

 

[Rick Darge]

Please someone explain to me why a guy like Roger Deakins who has shot Epic, prefers the Alexa for its sharp film like look? And then explain to me why Deakins' primary VFX friends are all vying for Alexa over the Epic? I mean.. these guys get paid millions to do what they do and they could shoot with ANYTHING they wanted. So why are they picking the 8-BIT HD Alexa as you are saying and not the 16BIT 5K EPIC? -- I'm sick of all the Versus debate. It's not all about how many features you can cram into a camera. The bigger the number isn't always the right way of doing things. And this idea of "future proofing" your work makes no sense to me. Just because studios and theaters and TV sets are slowly converting to UHD, doesn't mean Citizen Kane won't be popular anymore.. I totally get what certain people are saying, I'm one of them. It's always best to shoot with the cleanest image possible, yes, but I'm not sure anymore. We seem to just be leveling off. Everything looks the same. I'm cranky.

 

[PatrickFaith]

So tonight I was having a discussion with a friend of mine whose been a head at two different vfx teams that have won academy awards (he's honestly as professional as it gets in the post world of finishing) and the topic of resolution came up after Skyfalll was randomly mentioned. I opined that the film's lack of a greater resolution was actually felt in some scenes whereas I had never thought that with an Alexa film before. He countered that the EPIC would have only gave them 2.5k as opposed to 2k, "big deal."

I would generally agree with your fx friend, with some caveat's that many fx people calculate pixel resolution via the error in geometry solves and often see the file via a DI.

So what I think most people are getting as far as exact pixel resolution is:
2.7k on scarlet & redone (which is debayer plus the redcode at 7:1 which lowers organic spatial frequencies)
3.4k on the epic at redcode 9:1
4k on the epic at redcode 3:1
5k on the monochrome epic at 3:1

Most of the time, the fx guys see this through a DI, and sometimes people are even shooting through filters and not telling them ... so I can see how they could have a simplified view of the numbers.

 

[Christopher Barry]

And then explain to me why Deakins' primary VFX friends are all vying for Alexa over the Epic?

Putting the strive for technological perfection to the side, from a business and financial point of view, to some extent, compiling VFX at 2K (as opposed to 4K or 5K) to marry with the source footage of the Alexa at up to 2.8K may be more cost effective and profitable, perhaps one angle of consideration. That said, I am not addressing the 5K or 2.5K question, nor do I know the motivation of VFX producers in this suggested example. Intriguing question, Rick.

 

[David Mullen ASC]

So why are they picking the 8-BIT HD Alexa as you are saying and not the 16BIT 5K EPIC?

"Skyfall" was shot in 12-bit ARRIRAW.

 

[Blair S. Paulsen]

Considering the current financial challenges for the VFX industry, I would concur that the costs of working at higher resolutions is anathema.

IAC, picture quality for theatrical scale exhibition is a matrix of factors like acquisition resolution, bit depth, compression topology, grading/DI and, never to be forgotten - optics.

But lets get back to the OP's note about 2.5K vs 5K and the measurable resolution of a decoded Bayer CFA data set. For starters, every photo site - whether covered by a green, red or blue filter - creates a voltage value that the algorithm can use to affix luminance values and create edges. Yes, the blue and red records require more interpolation than the green record, but only counting the green channel is to ignore how the math works that underlies the function of a Bayer pattern reconstruction.

I would also point out that the OLPF filter and the design of the decoding algorithm are carefully coordinated. The obvious goal being to minimize aliasing while retaining resolution. Don't forget that the effect of the OLPF can be characterized and factored into the algorithm, its a known property. Properly utilized, the OLPF earns it's keep in the optical path by helping optimize usable resolution. Put another way, if aliasing artifacts are a major portion of your high frequency detail, then much of the the real detail can get lost in the pollution.

Cheers - #19

 

[Les Dittert]

I relish aliasing in my OLPF free shots, it tells me the lens is well focused and I will be able to reap higher than native resolution from the footage.
I would certainly not call it pollution, it is more like detail ripe for harvesting insane detail.
Future image processing will only improve on this, yielding even better results from the same raw footage I'm taking now. The advantage of RAW.
There is only so much detail that can be back calculated by deconvolution, unfortunately. Current OLPF is a bit 'wide' . They overshot in the safe direction.
I believe you will see less pixel spread ( not in microns, but pixels ) on the Dragon OLPF than current cameras. RXC will gain a more advanced debayer to help harness this pulling better detail as well..
Progress.

I would also point out that the OLPF filter and the design of the decoding algorithm are carefully coordinated. The obvious goal being to minimize aliasing while retaining resolution. Don't forget that the effect of the OLPF can be characterized and factored into the algorithm, its a known property. Properly utilized, the OLPF earns it's keep in the optical path by helping optimize usable resolution. Put another way, if aliasing artifacts are a major portion of your high frequency detail, then much of the the real detail can get lost in the pollution.

Cheers - #19

 

[Daniel Pearson]

While shooting with a stock Red can get about 3.5K, I go well beyond that.

Here is a shot that was shot 4K , Canon 17-55 f2.8 at F5.1. Handheld shot.
Deleted OLPF and special post processing is my modus operandi .

Go 6K

link to 6K jpeg, please load into capable image viewer to zoom 200%.

http://www.mediafire.com/view/?8hfytb52qjjgofb

Incredible detail in your shot Les, the work you're doing with your Scarlet appears to be paying off, but is all very curious.

 

[James_Mills]

The F65 is a completely different beast... We need the dragon before we want to start to comparing the epic with that one... lol.

Are you being sarcastic? Serious question...

Yes 5k is a larger sensor area than 4k, but it is 1/4 more sensor area, whereas 4k is only 1/3 more than 3k and 3k is 1/2 more than 2k, so there is a similar effect with all of these steps but to differing degrees.

So is the actual frame larger than the Scarlet 4k though? I thought they were both 35mm full frame size? Is the 5k epic frame the same size as a Canon 5d Mkii / mkiii then? If they are both 35mm size, then why would the 5k look better than the 4k if it's not cropping in on the sensor's actual image?

 

[Rick Darge]

"Skyfall" was shot in 12-bit ARRIRAW.

I know David, I was being a little sarcastic in my comments

 

[Noah Yuan-Vogel]

So is the actual frame larger than the Scarlet 4k though? I thought they were both 35mm full frame size? Is the 5k epic frame the same size as a Canon 5d Mkii / mkiii then? If they are both 35mm size, then why would the 5k look better than the 4k if it's not cropping in on the sensor's actual image?

You should really check out the data sheet threads that are stickied since this is very fundamental to all RED camera systems. Epic/Scarlet/R1MX use the same 5k sensor with the same pixel size so the sensor area is directly proportional to the resolution settings you are using. The whole sensor is capable of 5k in a sensor size between Full Frame 35 and Super 35, where 4k is about standard 35mm sized and 2k is about the size of standard 16mm. By comparison a 5D mk2/3 has about 2x the total sensor area as 5k RED MX (however its a bit more complicated when you consider active pixel area when shooting video).

 

[Vadim Bobkovsky]

I know David, I was being a little sarcastic in my comments

Roger Deakings himself wrote as an answer on his forum, that he very much used to the old-school optical viewfinder, so the camera without one (be it R1, Epic or Dragon) will be his least choice, really. He also mentioned, he would like to shoot with higher resolution camera, therefore Deakins would probably wait for Alexa 4K version. I don't expect Roger to get anywhere near the Dragon anytime soon, even though the dynamic range of new sensor may actually blow celluloid out of the water and I know from the interviews, that he often liked to overexpose with film.

Too bad, really, that Steven Soderbergh is retiring, apparently he was very excited about the Dragon sensor and I'd love him to shoot cinema or TV using it:

SODERBERGH: They’ve got a new sensor coming out, the Dragon, that’s fucking insane, like it’s a whole other level in terms of dynamic range, resolution, it’s crazy. I’ve seen the tests on it, it’s nuts. These guys just stay up really late and they’re just never satisfied and they’re constantly pushing it. So I look at it as…if you’re conservative you should probably use something other than the RED, if you’re someone who likes to live dangerously and push stuff then you should be using the RED because that’s where it’s at its best.
. . .
When do you think digital cameras are going to hit the resolution or the ability that anyone using film is going to say, “It’s time to switch?”
SODERBERGH: They’ve already done it. Film people are just in denial. They’ve already done it. They passed them a couple years ago.
Okay, let me ask you this question, when do you think digital cameras will get to doing what IMAX can do, if ever?
SODERBERGH: Oh, they will. I mean the new Dragon 6k is crazy. Take the new Dragon sensor, put a Master Prime on it, shoot it at a decent F-stop, and do a line pairs test where you have this chart that has a series of lines on it… and let’s see.

http://collider.com/steven-soderbergh-digital-cameras-48-fps-interview/

 

[James_Mills]

You should really check out the data sheet threads that are stickied since this is very fundamental to all RED camera systems. Epic/Scarlet/R1MX use the same 5k sensor with the same pixel size so the sensor area is directly proportional to the resolution settings you are using. The whole sensor is capable of 5k in a sensor size between Full Frame 35 and Super 35, where 4k is about standard 35mm sized and 2k is about the size of standard 16mm. By comparison a 5D mk2/3 has about 2x the total sensor area as 5k RED MX (however its a bit more complicated when you consider active pixel area when shooting video).

That's what I thought, I was just confused because I understood that you stated the 5k to 4k difference of 1k is technically tantamount to the 4k to 3k of the Scarlet's own cropping, however that doesn't make sense to me seeing that they're both using the same 35mm sized frame. If they're both using the same sensor and same frame size, I feel like the two images (at least on any monitor at 3k or lower) should be indistinguishable before blown up...?