Why 4k

[Graeme Nattress]

Even if you're thinking HD resolutions or 2k for RED, here's why I think 4k is the way:

1) Longevity: Get a nice shot in 4k and it will last years for stock photography. A programme made in 4k will be salable for many years to come.

2) Quality: There's no substitute for quality and that's what 4k gives you. On big screen, it's simply stuning.

3) Reframing: If you're doing an HD or 2k production, you've got lots of room to reframe and zoom with 4k.

4) Oversampling: Going down to HD? Now you're oversampling which is always great for image quality.

5) Unique Selling Point: Need to differentiate yourself from your competitors? Then shoot 4k when they're still stuck in HD.

Anyone have any more good 4k uses?

Graeme

[David Limpus]

Graeme what do you see as cost effective long term archive storage for the 4k data with passable transfer rates to and from the archive?

[Graeme Nattress]

Original 4k camera data can be kept in it's native REDCODE RAW form. That should keep data rates manageble to a great extent. The whole of the IT industry is working on ways to archive the gigabytes of data needed, and I personally think that normal hard drives will keep data ok until large format optical or holographic media is readily available, with data tape keeping us going through that.

Graeme

[David Limpus]

Is there a REDCode data rate table listed? With storage I will have to start look for a spectra T950 going cheap.

[Jim Arthurs]

4) Oversampling: Going down to HD? Now you're oversampling which is always great for image quality.

What I've said since the first images released is that the RED is the ultimate 2K camera due to the oversampling. Oversampling gives you desired natural sharpness, not artifical edge enhanced sharpness or an image that is overly soft, like I see from native 2K/1080 Bayer sensors.

Regards,

[Ace]

Oversampling apply directly to project.
Oversampling apply directly to project.
Oversampling apply directly to project.

[Blaine Golden]

I think this is just a no-brainer. Regarding storage: it's going to continue to grow and two years from now we'll be wondering how we ever got along with that measly 500GB hard drive.

[Michael Morlan]

This is good stuff. I've adapted it for my rental site. I re-wrote Graeme's initial thoughts as;

Longevity: A 4k frame can serve as stock and production photos. A 4K program will be sellable for many years to come.

Visual Effects: When the extra detail for a green screen key or difficult camera track is needed, 4K delivers four times the information as 2K.

Reframing: Producing an HD or 2K show? Use a 4K frame for re-frames and punch-ins.
Save time on the set by eliminating setups and performance discontinuities. DoP's like to re-light and reframe closer to the axis on close-ups, but when schedules are tight or a shot was missed, 4K can save the shoot.

Over-sampling: 4K down-sampled to 2K or HD produces more natural detail than native cameras' sharpening algorithms.


M

[Graeme Nattress]

2k downsampled from 4k can look nice, but, given a big screen and a projector, which is nicer:

4k on a 4k projector, or

2k downsampled from 4k on a 2k projector,

assuming both projectors are of equal quality.

Graeme

[Tonaci Tran]

Is this a quiz? 4K projection is inKredible and kills any 2k projector.

[Graeme Nattress]

It's not a quiz, but a philosophical question of sorts, mostly directed at Jim Arthur's post above.

Graeme

[Tonaci Tran]

Yeah, I hear ya Graeme. I think another important side-point to make is that RED will be able to capture beautiful 2k/1080p/720p natively even without downsampling. For my personal use, I will alyways try to shoot 4k at all times, however, there will be times where I'll want to shoot 120 fps in 2k.

[Graeme Nattress]

For 2k 120p you'd need an external recording device off the high speed port though. 2k 60p can go out REDCODE.

Graeme

[Tonaci Tran]

I read that for Miami Vice, they shot it mostly with the Viper, but had to resort back to film for some of the action sequences. I think having to go the external route for 120fps 2k is a small sacrifice compared to resorting to a film cam for anything higher than 60fps.

Anyways, with REDCODE there really isn't any other excuse not to shoot 4k unless you only have super 16mm/B4 lenses.

[Jim Arthurs]

2k downsampled from 4k can look nice, but, given a big screen and a projector, which is nicer:

4k on a 4k projector, or

2k downsampled from 4k on a 2k projector,

assuming both projectors are of equal quality.

Graeme

The 4K on 4K of course!

Well, then how about this... which will look nicer, starting from a Bayer sensor; 4K on a 4K projector, or downsampled 6 or 8K on a 4K projector?

Whenever possible, I simply prefer to think of post and finish at one level below whatever a camera is capable of.

This philosophy gives you breathing room for VFX and manipulation you simply wouldn't have if your final output is equal to your shooting size.

[Thomas Mathai]

Even if you're thinking HD resolutions or 2k for RED, here's why I think 4k is the way:

1) Longevity: Get a nice shot in 4k and it will last years for stock photography. A programme made in 4k will be salable for many years to come.

2) Quality: There's no substitute for quality and that's what 4k gives you. On big screen, it's simply stuning.

3) Reframing: If you're doing an HD or 2k production, you've got lots of room to reframe and zoom with 4k.

4) Oversampling: Going down to HD? Now you're oversampling which is always great for image quality.

5) Unique Selling Point: Need to differentiate yourself from your competitors? Then shoot 4k when they're still stuck in HD.

Anyone have any more good 4k uses?

Graeme

Of all the 4k DIs I have worked on, none of them have been shown theatrically in 4k yet. Who knows when they will be, because the reality is that not all theater chains are ready to convert to digital, let alone 4k.

4k to film I have noticed does look nicer, since it's a true film resolution.

Red 4k transfered to film is going to have to be downsampled since film 4k is limited to 4096x3112, with 440 pixels on the left for soundtrack, so 4520x2540, would have to be scaled to 3656x2054.

[Graeme Nattress]

I've not see that comparison - nearest I've seen is 65mm on 4k and RED, although having a very different aesthetic, (not just because it was shot with very intentions) RED didn't stand out as being necessarily being worse. The smaller grain on the larger film formats also help them look more like the RED footage.

[Thomas Mathai]

This is good stuff. I've adapted it for my rental site. I re-wrote Graeme's initial thoughts as;

Longevity: A 4k frame can serve as stock and production photos. A 4K program will be sellable for many years to come.

Visual Effects: When the extra detail for a green screen key or difficult camera track is needed, 4K delivers four times the information as 2K.

Reframing: Producing an HD or 2K show? Use a 4K frame for re-frames and punch-ins.
Save time on the set by eliminating setups and performance discontinuities. DoP's like to re-light and reframe closer to the axis on close-ups, but when schedules are tight or a shot was missed, 4K can save the shoot.

Over-sampling: 4K down-sampled to 2K or HD produces more natural detail than native cameras' sharpening algorithms.


M

Not many people are doing vfx in 4k because it's more than 4 times the work. I don't even think Weta, Pixar, ILM and Dreamworks are doing everything in 4k, and they have more rendering power than any of us.

The reality of greenscren/bluesreen work is that more often than not, you need a good roto team to fix things. Rotoing in 4k is going to take a long time, no matter how good you are.

CG work in 4k is just going to be nuts even with render farms.

GPU rendering will help, but that's still mostly for previewing.

[Graeme Nattress]

I don't see 4k as being 4 times the work though,

4k rotoing isn't going to be slower. Roto work is a human tracing an edge with vector tools. Image resolution doesn't make this process take longer, and the cleaner the edges to see, the better, I'd think, not worse.

CG work is going to take longer to render. That's true, but it's not the end of the world as render power is growing much faster than image raster size :-)

That said, I see the biggest 4k slowdowns are the things you do every shot - colour correction, and that's where a good GPU can really help, and that technology is with us today.

Graeme

[Alex Boothby]

Greame: I think I once heard you say that REDCODE is only for footage coming out of the camera. But it strikes me that REDCODE would be a fabulous archiving format. At the moment I plan to convert all RED 4K REDCODE footage to tiff image sequences (uncompressed) for post processing, grading, vfx, etc. Is there a way that we can re-compress these final shots to REDCODE (on redcine?) at the end of the day before archiving?

Thanks

[JW Lee]

Is there a benefit in terms of controlling depth of field with 4k over 2k? Can I get a shallower DOF with 4k? If so, how substantial is the difference.

[Michael Morlan]

Greame: I think I once heard you say that REDCODE is only for footage coming out of the camera. But it strikes me that REDCODE would be a fabulous archiving format. At the moment I plan to convert all RED 4K REDCODE footage to tiff image sequences (uncompressed) for post processing, grading, vfx, etc. Is there a way that we can re-compress these final shots to REDCODE (on redcine?) at the end of the day before archiving?

Thanks

Wow! Uncompressed TIFF's are hardcore! RLE, zip, rar, and tape hardware compression do a nice job of reducing the size of TIFF's and are more of an IT industry standard.

No need for REDCODE.

M

[Jim Arthurs]

I don't see 4k as being 4 times the work though,

4k rotoing isn't going to be slower. Roto work is a human tracing an edge with vector tools. Image resolution doesn't make this process take longer, and the cleaner the edges to see, the better, I'd think, not worse.

CG work is going to take longer to render. That's true, but it's not the end of the world as render power is growing much faster than image raster size :-)

That said, I see the biggest 4k slowdowns are the things you do every shot - colour correction, and that's where a good GPU can really help, and that technology is with us today.

Graeme

Hi Graeme... this is an excellent conversation and I'm enjoying the dialog. I fully intend to monopolize a few minutes of your time at NAB and a bit longer block when I come out to Cali to pick up my RED in person. :)

First, I haven't seen but two displays of single projector 4K... the CineAlta Festival two NAB's ago when the Sony projector was just cutting its teeth (showing all 1080 material if I remember) and last NAB where I saw the Mystic India footage that was from 65mm. I haven't had the chance to see RED footage yet... I hope the upcoming NAB takes care of that!

My guess is that a larger than 4K sensor will be needed to pack a 4K final projection frame as well as the current 4K RED design will pack a 2K projection frame. But God Bless the RED for starting out with all those pixels.

But on to the CGI aspect of this thread... Thomas has excellent points, and I want to add my work experience into this.

First, 4K roto will be much more difficult because the artist precision needed is that much higher. Hair that's a simple blur in SD or HD resolves into individual strands at 4K. I've had personal experience on feature films doing roto (sadly) and you can get away with murder at SD compared to HD or 2K.

Second, CGI 4K effort won't be a simple factor of render power applied. God I wish it was. Let me address some problems I ran into on a 4K by 4K job, completely ignoring all issues related to computer power or memory requirements...

1.) Texture maps. Even for SD work, 2K texture maps are often used. In doing FX at 2K I've often had to paint or stitch up images at 10K or greater. To make 4K look GOOD you need MUCH larger imaging ability... beyond what you can easily get in one snap of the best DSLR today. Again, labor and time. The resolution required for an environment pano for a 4K scene using a CGI camera of moderate focal length boggles the mind. At least my feeble mind.

2.) Model quality. Doing a quality CGI model that will hold its close-up in 4K becomes that much harder and takes more man hours. Again, this is application of human labor.

3.) Aliasing control. In CGI you often have to render at a size bigger than your final to keep certain things from aliasing (fine lines, regular patterns). How big will you have to go to "fix" these issues if your nominal size is 4K? Big, bigger...

These are a couple reasons on top of all the other factors previously mentioned that keep 4K FX from being commonplace. You haven't seen a high shot count movie be all 4K for very good reason.

Of course all this will change in time, and at a fairly fast pace, I imagine. But now, TODAY, I'm very pragmatic and realistic when thinking about projects, and a 2K finish starting from the 4K RED "negative" makes the most sense for the kind of work I'm interested in.

On the flip-side, I'd gladly post a "bunch of people in a diner talking" with minimal fx sort of movie at 4K however... :)

[Alex Boothby]

I think Spiderman 2 was all 4K.

But I agree. The more post work you do, the more it makes sense (currently) to shoot 4K and finish in 2K or HD. Or maybe I'm just a wimp.

[Thomas Mathai]

I don't see 4k as being 4 times the work though,

4k rotoing isn't going to be slower. Roto work is a human tracing an edge with vector tools. Image resolution doesn't make this process take longer, and the cleaner the edges to see, the better, I'd think, not worse.

CG work is going to take longer to render. That's true, but it's not the end of the world as render power is growing much faster than image raster size :-)

That said, I see the biggest 4k slowdowns are the things you do every shot - colour correction, and that's where a good GPU can really help, and that technology is with us today.

Graeme

I disagree with 4k roto not being slower. OF all the roto work I have done in the past, it was easier get through it in SD than 2k. I would have more resolution to find finer edges, but overall it took longer.

When CG renders take 24 hours per frame at 2k from what I have been told at Siggraph presentations, I see it being even more complicated when you have to render out at 4k.

It's not just a matter of switching out a quad core machine with an 8 or 16 core machine either. A lot of it is knowing how to manage and streamline your rendering.

[Thomas Mathai]

I think Spiderman 2 was all 4K.

But I agree. The more post work you do, the more it makes sense (currently) to shoot 4K and finish in 2K or HD. Or maybe I'm just a wimp.


Spiderman 2 was a 4k DI, but all the visual effects work was done in 2k.


With Flame/Inferno, you can use proxies to do your roto, paint and comp work in a lower resolution, then it would automatically apply those results to the higher resolution source.

That should be how we should have it in the future, doesn't matter what your source is, allow you to work in proxy mode, and then allow you to render at whatever resolution you decide, yet you have the project files to go back to later when you want to rerender at a higher resolution.

[Emanuel A.]

I don't see 4k as being 4 times the work thoughBut 4K is 400% more (2K compared) though.

[Brook Willard]

But 4K is 400% more (2K compared) though.

No, it's 300% more. ;)

[Emanuel A.]

No, it's 300% more. ;)?

Well,

[@16:9 ratio]
2048 x 1152 = 2 359 296 [2K]
4096 x 2304 = 9 437 184 [4K]

it gives 400% more...or am I wrong or are you joking with me, Brook? ;)

[Ruairi Robinson]

?

Well,

[@16:9 ratio]
2048 x 1152 = 2 359 296 [2K]
4096 x 2304 = 9 437 184 [4K]

it gives 400% more...or am I wrong or are you joking with me, Brook? ;)


100+300=400

[Gavin Greenwalt]

Render times can be > 4x because of ram limitations and how you break up the BSP trees. It'll also require everything to be detailed better, not just rendered at a higher resolution, that means more time.

I agree with Roto taking longer as well. with DV you can get away with a lot of cheating, not so with 4k. If you actually need to 'roto' 'roto' and not just key the hair, you have more hair you have to roto in 4k.

[Graeme Nattress]

CGI rendering is very parallelizable though. For texture complexity, hybrid texture + procedural shader will give all the detail you'll ever need :-)

If you're going to to bother of rotoing every hair visible in a 4k shot, methinks you've got other problems.... Perhaps that shot is better done all CGI, or... ?

Graeme

[Thomas Mathai]

CGI rendering is very parallelizable though. For texture complexity, hybrid texture + procedural shader will give all the detail you'll ever need :-)

If you're going to to bother of rotoing every hair visible in a 4k shot, methinks you've got other problems.... Perhaps that shot is better done all CGI, or... ?

Graeme

The realities of production is that not everything is thought out or expected to be an effects shot.

I once had to roto, in 2k, a man swinging his body around wildly, of course he had long hair. It wasn't meant to be an fx shot, but the production decided to digitally change the background.

[Graeme Nattress]

Dratted people who decide to change backgrounds :-) I don't envy you on that one!

Graeme

[Tonaci Tran]

I found this quote amusing from one of the colorspace guys regarding 4k:
http://www.dvxuser.com/V6/showpost.php?p=557920&postcount=34

"As far as 4k versus 2k, I really don't want to turn this into a long rant, but if you know even a single person who is working 4k, I would be very suprised. If fact, i'd be [a bit] suprised if you knew someone who even was working with 4k images from a D-SLR, since there are maybe 8 D-SLRs that support this resolution... and D-SLRs are meant to approximate more than double the resolution of 35mm motion picture film.

The fact of the matter is that 4k is an exorbitantly expensive workflow to support; and packing that type of resolution into a 35mm chip degrades the type of performance you can see in other areas. We considered 4k, but 2k is the current standard for the industry, and that's why we chose 2k-- it ended up being the "sweet spot" with regard to the total performance of our chip. I'm not saying 4k isn't great, or that the industry won't eventually shift there, but that isn't the current standard, and going 4k would have required us to make sacrifices in other areas. "

A quote from David Stump regarding 4k back in November:
http://www.assimilateinc.com/pdfs/Q&A-DavidStump.pdf

What do you like about 4k?
One word: clarity. It harkens back to the beauty you saw achieved in the first print from 65mm or Vista Vision - strikingly crisp and clear.
I expect the clarity and resolution of the Red 4k camera to be just as good. The results are so striking that studios will now be willing to take the risks to achieve more resolution, more color space.

[Graeme Nattress]

Thanks for that Tonaci - most amusing. I don't think anyone who's seen what 4k can do would want anything less :-)

Graeme

[JW Lee]

bump.....
How does 4K impact DOF and how does it compare with 2K?

[Graeme Nattress]

Well, technically, it doesn't. Only thing that affects DOF is aperture. However, larger sensors make longer lenses look shorter, and longer lenses magnify the image making blurred background appear more blurred, hence make DOF look shallower. So, simple answer is 4k looks shallower, but technically nit picking answer is that it doesn't change a thing.

Graeme

[Chris Gearhart]

Mike Curtis had a link at HDfordindies on DOF.

http://www.film-and-video.com/dofmyth.htm

[Júlio Taubkin]

Well, technically, it doesn't. Only thing that affects DOF is aperture. However, larger sensors make longer lenses look shorter, and longer lenses magnify the image making blurred background appear more blurred, hence make DOF look shallower. So, simple answer is 4k looks shallower, but technically nit picking answer is that it doesn't change a thing.

Graeme

Well, isn't DOF dependent of the Circle of Confusion used? A 4Kprojection has a different CoC of a 2K projection. So Wouldn't it have some consequences on DOF?

(For instance, 35mm motion picture has different DOF when it's projected in a theather or when it goes to SD television. Technically, the DOF is the same, but since there aredifferent tolerances for the different viewing methods, DOF does change...)

[Gavin Greenwalt]

Well, isn't DOF dependent of the Circle of Confusion used? A 4Kprojection has a different CoC of a 2K projection. So Wouldn't it have some consequences on DOF?

(For instance, 35mm motion picture has different DOF when it's projected in a theather or when it goes to SD television. Technically, the DOF is the same, but since there aredifferent tolerances for the different viewing methods, DOF does change...)

The CoC will change, but it'll be a linear shift so it won't effect the DoF just the 'focus'. It's like applying a Gaussian blur to a 4k image. The DOF hasn't changed, since everything was affected evenly.

[Jeff Kilgroe]

I once had to roto, in 2k, a man swinging his body around wildly, of course he had long hair. It wasn't meant to be an fx shot, but the production decided to digitally change the background.

<warning, stupid question> ..Uh, why couldn't the shot just be re-done?

I know, I know... it doesn't always work that way, especially when the decision is made weeks or months down the road. I feel your pain though. :cool:

[Jeff Kilgroe]

I don't see 4k as being 4 times the work though,

4k rotoing isn't going to be slower. Roto work is a human tracing an edge with vector tools. Image resolution doesn't make this process take longer, and the cleaner the edges to see, the better, I'd think, not worse.

I'm with Graeme on this one. And if the milk girls are anything to judge by, my roto and keying work with RED will become a much better experience. It's all about the crispness and clarity of the imagery you're working with and I would gladly do comp work with 2K or 4K RAW from a good camera system with great optics any day over 1080i DVCPROHD or <oh, lord!> HDV.


CG work is going to take longer to render. That's true, but it's not the end of the world as render power is growing much faster than image raster size :-)

Yep. :-) I've got a decent render farm -- puny by Hollywood standards, but I can render faster than I (and my limited help) can work. I often wonder how many "pro" 3D guys out there get by without a decent render farm, even single-man operations. With a good farm, I can take on more jobs at lower cost, faster turn-around times, I can have rendered sequences often while working direct with the client on changes, etc.. I've just made it a habit from the beginning to allways include the cost of a new render node into a job wherever I can. ;)

[_BK]

I don't see 4k as being 4 times the work though,
4k rotoing isn't going to be slower. Roto work is a human tracing an edge with vector tools. Image resolution doesn't make this process take longer, and the cleaner the edges to see, the better, I'd think, not worse.


Interesting thread here.... I dont think rotoing 4k will be 4x as slow, however it will be quite a bit slower than 2k or HD. When doing roto you need to push an enormous amount of data through a network as the roto artist is constantly scrubbing frames and caching frames into memory. At 4k the artist will have to still push 4x the data though the Roto package and cache. Sure, you can do most in proxy res, but final roto would still have to be done at the full res.

Im scared to think about doing closeup CGI at 4k, the amount of work and rendering power/memory needed to do that is unreal. But, the availability of a 4k plate even if you are outputting at 2k or HD is great for things like tracking and matte paintings.

[Zak Forsman]

this freedom to reframe in post is very intriguing. i do alot of improvisation so the ability to punch-in and keep the continuity of the performance is... wow. i've got to think about this. suppose i'd have to settle on mastering out to 2K. at 4K the punch-ins and reframing would be pretty obvious, im sure.

[Graeme Nattress]

Aparrant DOF is dependent on a lot of factors, including sensor size and size of the final image. Technically DOF is constant with those factors, but the human visual perception system can be made to think the DOF has changed.

Graeme

[Martin Drew]

Mike Curtis had a link at HDfordindies on DOF.

http://www.film-and-video.com/dofmyth.htm

I had seen Walters site ages ago and the DOF myth argument has always bothered me. The focus of a lens is an absolute distance, the DOF is the range of distances either side of that focus distance which will give acceptable focus and as such image magnification is one of the parameters to determine DOF. The example he gives someway down the page illustrates the point. These 2 images are taken with lenses of different focal lengths.

http://www.film-and-video.com/images/camclose80mm.png
http://www.film-and-video.com/images/camclose40mm.png

he argues that while the DOF may appear different it is in fact an illusion because if you match the size of an out of focus item in the image between the 2 images you will see that there is the same amount of blurryness on each image. The problem is that if you are magnifying a part of the image you are changing a DOF parameter.

Or am I missing something?

Martin

[Júlio Taubkin]

I agree with Martin. DOF is not about what's in focus or not. It's about how far from focus we will still accept and perceive as "in focus". As such, it's dependent of a lot of factors, especially size of final image (or our ability to perceive the full resolution of it). If you take a 4K image and downrez to ntsc, you will gain DOF, because it will be more difficult to realize the small nuances of your DOF. That's why I think that a 2K downrezz image should have a different DOF.

[Gavin Greenwalt]

<warning, stupid question> ..Uh, why couldn't the shot just be re-done?

I know, I know... it doesn't always work that way, especially when the decision is made weeks or months down the road. I feel your pain though. :cool:

Sky Replacement is a popular 'Fix it in Post' task which often requires the footage to be shot in a way not conducive to a standard FX shoot. And it's usually even decided before they shoot it that way. So it's not so much an "error" as a challenge.

it's also the reason I always take a picture of a good sky when I see it. You never know when it might come in handy later.

[Martin Drew]

The CoC will change, but it'll be a linear shift so it won't effect the DoF just the 'focus'. It's like applying a Gaussian blur to a 4k image. The DOF hasn't changed, since everything was affected evenly.

If the circle of confusion changes it will affect the depth of field. circle of confusion is one of the parameters used to define depth of field.

Martin

[Hans von Sonntag]

My first Post in this forum so please be patient...

I like to add some practical thoughts to this very interesting 4K vs 2K thread:

One of REDs major advantages over a 35mm camera is not only costs but the possibility to use 35mm and S16 DOF in one camera. For example:

For exterior shots at night it is nice to use available lights and some practicals among extra light-equipment. Therefore I don't need only fast stock (RED is supposed to be 400 ASA as far as I know which is considerably fast even compared to 500 ASA Kodak/Fuji. You will only rarely expose this stock to its full 500 ASA) but also fast primes, e.g. T1.4. This will lead to very shallow focus and on longer shots to a soft, undefined picture. Not necessarily nice. Now I window the sensor to S16 and the focus will be not as shallow. Great.

Next example: I use a short zoom lens, a 5 to 1 Cooke 18-100 for instance - the only lens I have with me. For a certain shot a longer lens would be nice. No, I'm not going to use this extender. It eats 2 stops and messes up the sharpness of the lens. But - I've got RED. I just simply window the sensor and have an almost 2 times closer image. Great.

To come back to the 2k vs 4k business. As long as any decent DI software needs uncompressed video (QT, DPX, etc.) I need a fast raid and a lot of storage to manage even HD. 2K will be possible; I might have to by a new raid though (another 10k). 4K is not possible. Not now, not tomorrow, not for an indie filmmaker in the next 2 years or so.
If one once tackled messy keys (real world greenbox shots are always kind of messy) and rendered them and rerendered them (you hardly render a complex composit only once) knows how long it takes on HD. 4K? Please not. I'm not going to do it.

So dear RED Wizards: Please make REDCODE RAW possible not only for windowed S16 but also 35mm full aperture. I want to shoot all in 2K and just switch the aperture but not the codec. Another plus: I can overcrank the camera up to 60fps which is enough for the most circumstances. 4K is great for the future and its good to know that the investment will hold its value for a longer time but 2K is much more realistic for an indie filmaker to day (and I am doing my stuff to day and not in 2 years). This actually reminds me of the days when I upgraded my old Arri SR from N16 to S16 which added another 10 year of usage.

Hans

[Graeme Nattress]

REDCODE RAW works in 4k and 2k modes, so you can leverage it for much cheaper 4k workflows with it's mild, great looking compression.

Graeme

[Hans von Sonntag]

Graeme, thanks a lot for your reply!

I suppose I did not understand your chart entirely.

So 2K REDCODE RAW does also work with a non-windowed sensor?

Is REDCODE RAW 4K/2K natively editable in NLE (e.g. FCP) right from the fly?

Does REDCine work like a conforming-tool? Can I put in my 4k/2K REDCODE RAW camera-files and conform via EDL/XML the edited shots into say 2K DPX/ 1080P_QT?

Please excuse me asking questions which might have been answered already elsewhere...

[Graeme Nattress]

I think the chart was done before we announced 4k RAW. You can have 4k RAW and 2k RAW both in REDCODE RAW.

Any app that supports Quicktime should edit it straight away.

There is some sort of conform type functionality in REDCINE, and I think you'll find more info on that elsewhere on the forum.

Graeme

[Hans von Sonntag]

Thank's a lot! Much appreciated,

Hans

[Antoine Baumann]

One of REDs major advantages over a 35mm camera is not only costs but the possibility to use 35mm and S16 DOF in one camera.

I totaly agree with that. I love the possibility to switch between the two format. Also the 2k shooting allow greater frame rates, so for some shots it would be necessary to shoot 2k (you could even need to go down to 720p for 120 fps). That is an other reason that would make you go for a 2k postproduction.



To come back to the 2k vs 4k business. As long as any decent DI software needs uncompressed video (QT, DPX, etc.) I need a fast raid and a lot of storage to manage even HD. 2K will be possible; I might have to by a new raid though (another 10k). 4K is not possible. Not now, not tomorrow, not for an indie filmmaker in the next 2 years or so.
If one once tackled messy keys (real world greenbox shots are always kind of messy) and rendered them and rerendered them (you hardly render a complex composit only once) knows how long it takes on HD. 4K? Please not. I'm not going to do it.

I think you are totaly right about the raid system, and the number of time you need to render a comp, thus the time it would take at 4k, but I am personnaly thinking to stick to redcode rgb for the postproduction work, as long as I can (might be even for color correction). And as redcode is a quicktime codec, it will work in all the softwares supporting them (FCP, Shake, etc...).
I do not see the point (but I am open minded) to use uncompressed tiff, if you shoot redcode. It will be much lighter to use redcode and will save storages acquisition (for exemple the milkgirls 1080 redcode is only around 7.5 MB per second), if the software you use support it of course.

My 2 cents,
antoine

[Nick Shaw]

…for exemple the milkgirls 1080 redcode is only around 7.5 MB per second…

7.5MB/s is the data-rate for the MJPEG 1080p Quicktime of the clip available for download, not the datarate of the original REDCODE clip. REDCODE at 24fps is expected to be about 27MB/s (although Graeme has recently said they have improved efficiency since that figure was quoted.)

Nonetheless, I agree that REDCODE will make working with 4k media much more feasible on a 'normal' editing system.

Nick

[Rob Lohman]

The 1080 milkgirls clip was not in REDCODE. Otherwise you would've had to install our codec to play it.

p.s. the 27 MB/s number was for REDCODE RAW, not RGB. That might be higher (RGB has 3 times as much data than RAW).

[Hans von Sonntag]

REDCODE. Ah I missed something, I suppose. I forgot in my thoughts the possibility to use REDCODE RGB for postproduction which seems to be something similar to that what Cineform is already doing and which obviously is working fine. It will be possible to work with a lot "lighter" files, than for instance uncompressed TIFF, without the downside of non-acceptable compression/colourspace (HDCAM, DVCPROHD, not to mention HDV) which is really great.

Nonetheless rendering a 4K comp to REDCODE will be the same time than rendering to a TIFF-sequence.

A fail-safe Raid will be still necessary in an REDCODE workflow, especially if one thinks about data backup/safety - no 35mm negative to rely upon just some data which can vaporize in a second.

I must say Antoine is right pointing out the overcranking possibilities in the 2K-mode.Yes, 4K is super-great quality-wise but if you want to use the camera in the most flexible and creative mode 2K seems the way to go.

Hans

[Rob Lohman]

Rendering times will not be the same for REDCODE vs. TIFF if you have a fast enough RAID setup. If the wait is not on writing the data it's on compressing it. Since REDCODE is wavelet based it requires time to compress versus the non-compressed (normally) nature of TIFF.

The footage I shoot is always immediately backup up onto 2 systems. It would be wise to have at least 1 copy off-site. Then you make a third copy to work on and edit.

For working with high-res footage a fast array will be a must. Werther you also want to include safety in that RAID depends on a couple of things, including money.

As long as my original negatives + NLE projects are safe I can withstand a failure of the RAID array containing the footage.

[Holosynthetic]

Would someone be willing to talk about uncompressed 4.5k? like what's its advantage over compressed 4k besides the obvious, more resolution.

[Antoine Baumann]

Well uncompressed means there is all the informations that the camera captures. That means that this is the ultimate quality your camera could get. And if you are going for compositing work, you start with the best quality possible, knowing that any image transformation is going to destruct some informations. In uncompress world every info is there, in compress, well you take out some infos (that you think is not visible, or very hardly) and get smaller files.
But there is a multitude of kind of compression, and redcode seems to me pretty good visualy. I mean what is the point (again I am open minded) to shoot uncompressed, if on the final product, no one can make the difference, with a clever compression? But I am not saying that you cannot see any difference between redcode and uncompress (neither am I saying that you can see it, anyway the red cam and redcode are not finished yet).

Uncompressed 4,5 k is going to be something huge to manage, think that it will be in raw 10 bit 4520x2540x10x1x24 = around 2750Mb/sec (@ 24 fps) making around 1.225 TB/hour....and when you process them in rgb 12 bit, it will be 4520x2540x12x3x24 = around 10Gb/sec (@ 24 fps) (I guess this one would be hard to play real time, or very expensive if you prefer), making around 4.5 TB/hour.
Unless you are planning to invest a lot of money, I think you cannot really manage a total 4.5 k uncompressed workflow, unless you are doing very small project, such as advertising, music clip, or short movies.

My 2 cents,
antoine.

[Rob Lohman]

The main advantage in my opinion is higher framerates. 4.5K up to 60 fps & 2K up to 120 fps.

Otherwise REDCODE compression should be good enough for most.

[Rogelio Salinas]

1st Ever Post. I just reserved my RED yesterday and a 18-85mm lens. Praise God. Will REDCODE be able to capture 4K at 60P or this only when you capture uncompressed? I believe that REDCODE can attain 60P, but only in 2K. Is it possible to shoot 24P in 4K and 60P in 2K and edit them together on the same project with REDCINE?

[Rob Lohman]

REDCINE is not an edit application. It is a RAW conversion like tool for moving images. It allows you to convert the camera format to whatever you want to edit with and get the color in the ballpark etc.

Obviously REDCINE will load the camera generated formats.

The 60 fps in 4(.5)K is only available through the RAW port.

[Andrew M.]

Rob, do you know by any chance how many photosensors MYSTERIUM CMOS has? (not pixels)

Andrew

[Rob Lohman]

twelve

[Graeme Nattress]

It has in total 4900x2580 photosites.

Graeme

[Andrew M.]

Let me expand my question since I think I didn't express myself clearly.

Pixel, is defined as all elements (photo sensors) required to capture white color representation of the display or capture point.

Also if MYSTERIUM has Bayer or any other green color enhancer it should have two photo sensors for green or a bit larger sensor for the green color.

Thus, RGB and one more G this gives you 4 photo sensors.
Some CMOS have all RGB sensors stuck on top of each other but I will count it as a 3 photo sensors anyway.

So again, how many photo sensors we have sensing independently the light, regardless of the color, coming to the MYSTERIUM CMOS chip?

Andrew

[Graeme Nattress]

We've discussed bayer CFA technology at length on this and the DVXUser and DVInfo forums. The Mysterium sensor has 4900x2580 photosites, and it uses a standard bayer CFA.

Graeme

[Antoine Baumann]

Hello Rogelio,

I think 4k @ 60fps is only possible with uncompressed, because redcod cannot process so quickly.
The only way I think you could edit 4k images and 2k images on the same project is to downsize your 4k footage to 2k in redcine, or you might be try to mix scan from 4k and 2k when print back your images on film. This might work as well (but only if you are going back to film).

antoine.

[Mark L. Pederson]

The 60 fps in 4(.5)K is only available through the RAW port.

Rob, my math is rusty - what is the write speed you need to capture 60fps 4.5K RAW ??

[Rogelio Salinas]

REDCINE is not an edit application. It is a RAW conversion like tool for moving images. It allows you to convert the camera format to whatever you want to edit with and get the color in the ballpark etc.

Obviously REDCINE will load the camera generated formats.

The 60 fps in 4(.5)K is only available through the RAW port.

How exactly would you capture the footage through the RAW port and get it into the CPU? I know you would run it through REDCINE, but what is the best method to do this?

[Rob Lohman]

RED-RAID connects to the RAW port and takes care of all of this. This is a product we'll launch later so no specific details are available yet. The maximum datarate over that port will reach around 900 MB/s (that's megabytes a second).

[Andrew M.]

We've discussed bayer CFA technology at length on this and the DVXUser and DVInfo forums. The Mysterium sensor has 4900x2580 photosites, and it uses a standard bayer CFA.

Graeme

Graeme, sorry for this but I went to DVXUser forum and searched by keyword MYSTERIUM and spent 2 hours reading all posts and nothing about how many pixels MYSTERIUM has. Also I did try keyword CFA and not even one post came out there.

Just wanted to explain the confusion, since I have seen many references to the MYSTERIUM as a 12 Mpixel sensor but if it has 12M photo sites then it is only 3 Mpixel sensor, correct?
Otherwise MYSTERIUM have to have 48M photo sites (photo sensors).

Could you be so kind and try to find the link to the discussion about it on the other forums, please.

Andrew

[Graeme Nattress]

Andrew, we've talked at great length about the bayer sensor. To say that it only has 3 million pixels is totally wrong. The sensor has 12 million or so photosites. When you run that information through a high quality demosaicing algorithm as we do, you get 12 million RGB pixels out, each different and unique, and you get a beautiful image.

The sensor is just that, a sensor - not a camera. A camera is a system of lens, sensor, electronics and computer software. You've got to look at the light that goes into the lens, and the image you get out the other end.

Graeme

[Nick Shaw]

The Mysterium is referred to as a 12MP sensor in the same way as the Bayer sensors in D-SLRs have their resolution in megapixels quoted based on the total number of photosites, not the number of Bayer 'clusters of four'. As Graeme has said many times, good de-Bayer algorithms will produce a much higher RGB resolution than one pixel per Bayer cluster (>75% RAW resolution.)

Nick

EDIT
OK Graeme beat me to it and answered while I was typing

[Graeme Nattress]

If you look at a bayer CFA, and just look at one pixel, you can see that each pixel has it's own neighbours that make up the 2x2 repeating bayer pattern. Each overlaps with the other. So, there are (ignoring the border condition) as many 2x2 bayer cells as their are pixels.

Bayer sensors are also much more efficient than the 3CCD or foveon approach. In those so called RGB sensor approaches, you get, in terms of RGB pixels, photosites/3 resolution out. With a bayer pattern sensor, you get RGB pixels = photosites out, but they're a very nicely interpolated instead of sampled directly. Tests have shown that, say, a 12mp bayer sensor is quite superior to a 6mp RGB style sensor in terms of resolution. Given that RGB sensors on stills cameras omit the vital anti-aliasing filter, so that they tend to look horrible, that conversion factor is hyper-conservative, but let's go with it for now. So, the bayer gives > 6mp full, real, resolution (looks much nicer though, and is ready and nice to look at 12mp) but to get 6mp resolution from a RGB sensor needs 3x the number of photosites, or around 18mp. So, for the quality we get from 12mp, an RGB sensor needs at least, probably a lot more than 18mp of photosites. That's inefficient as they don't exploit any spatial offset. And as soon as you try to interpolate that RGB image up to the size of the bayer image, you'll get nasty interpolation artifacts, and if it's a stills camera, they'll be amplified by the lack of necessary optical low pass anti-alias filtering.

Next, you need to know that RGB sensors don't come as big as Mysterium, and certainly have no where near the necessary frame rate. Next, you need to know that they're also heavily interpolated! Yes. That's because the colour filters they use are poor, and the red, green and blue are all mixed together, and seperating them out requires a fair bit of processing. So much for not being interpolated! The colours direct off the Mysterium are already very close to what they should be and require relatively minor correction to get them looking spot on.

Then, you need to know that the 3CCD approach doesn't work beyond a certain size, and that size is a lot smaller than super 35mm film. And, that the prism causes it's own optical nasties making chromatic aberation appear worse due to the different focal lengths needed for the red, green and blue sensors.

So, you can see there's a reason we do things the way we do them :-) And the results speak for themselves - lovely!

Graeme

[Andrew M.]

If you look at a bayer CFA, and just look at one pixel, you can see that each pixel has it's own neighbours that make up the 2x2 repeating bayer pattern. Each overlaps with the other. So, there are (ignoring the border condition) as many 2x2 bayer cells as their are pixels.
Graeme


Graeme, above explanation I like the best, in reality Bayer CFA pattern on 12M sensors CMOS has in reality 9 RGB Mpixels that are completley unique and it is the worse case scenario.

Thanks! a lot!!!

BTW I found the link you were talking about, on dvinfo forum
http://dvinfo.net/conf/showthread.php?t=80852&page=2&highlight=MYSTERIUM

Andrew

[Andrew M.]

Good explanation of Bayer Color Filter Array to RGB converssion...

http://www.helicontech.co.il/whitepapers/bayer-rgb.html

Andrew

[Rogelio Salinas]

Hello Rogelio,

I think 4k @ 60fps is only possible with uncompressed, because redcod cannot process so quickly.
The only way I think you could edit 4k images and 2k images on the same project is to downsize your 4k footage to 2k in redcine).

antoine.

Thanks for the update Antoine. Do you think there would be a noticable difference if all the 24P was shot in 4K and the 60P was shot in 2K once it is all resized to 2K or less? I would prefer to shoot everything for a feature in 4K, but if 4K 60P is going to be too an expensive alternative, then I would rather just shoot the 60P footage in 2K. My concern is just that there could be a noticable difference in the footage once it has been resized to either 1080P, 720P, or 480P. If not, then I may just shoot 24P in 4K and 60P in 2K.

[Antoine Baumann]

Well, first of all I have to say that I don't know, as I never shot 4k images and downsize them :-) therefore it is just a guess.

I think that the smaller will be your output (let say 480p) the less difference you could possibly see between a 4k original or a 2k, and the bigger the output, the more difference ( I know this one is so true, it is useless :-). But I think it would be ok at least up to 720p, may be all the way up to 2k...we will have to try.
I am definitely going to run such test, as soon as I get my red one.

ciao,
antoine

[Bruce Allen]

Also, Bayer works more like the way the human eye works - eg our luma resolution is higher than our chroma res...

So perceptually to humans, a Bayer 12mp image > a RGB 6mp image.

One problem is that After Effects, etc don't work in Bayer space or any space other than 4:4:4 RGB. Eg. you can't get a Gaussian blur that works at high res on the luma info but at lower res on the chroma info.

So, 4k renders suck. I just finished some stuff at that res for a trailer and it was not fun. By the way, bear in mind that 95% of trailers you see were finished at 2k or even HDCAM! Do they look soft? Of course not! So even if you shoot 4k, consider going to 2k as soon as possible. Anything more is arrogance.

3k 4:4:4 or 4k with Bayer color space is a decent compromise, of course, except the latter option as mentioned above isn't availble from an image processing perspective.

Graeme, care to write some RAW-space optimized plugins in your spare time? ;)

Bruce

[Jim Arthurs]

So, 4k renders suck. I just finished some stuff at that res for a trailer and it was not fun. By the way, bear in mind that 95% of trailers you see were finished at 2k or even HDCAM! Do they look soft? Of course not! So even if you shoot 4k, consider going to 2k as soon as possible. Anything more is arrogance.

My feelings as well, which I expressed early on in this thread. Those of us who have had to actually DO 4K or greater rez animation and effects know that it's more than sheer additional render power, that it's a level of LABOR and TIME for creation beyond what's acceptable at 2K.

Point to ANY major feature film in production or pre-production today with a significant volume of VFX that has a complete 4K post workflow... I don't think you can.

Regards,

[Andrew M.]

I spent few hours updating myself on the CCD and CMOS sensors.
I see industry is in urgent need to redefine definition of the pixel.
Different for displays different for sensors etc.

The best will be to refer to RGB pixel as a 'MP' and for total count of photo-sites a 'Mp' and then what we do with Bayer CFA pixels? MpBCFA?

Like we have done it for Bytes MB (bytes) and Mb (bits)

So my conclusion is, MYSTERIUM is 12MpBCFA sensor.
I guess you Guys new it all along....
Andrew

[Bruce Allen]

Andrew... you're saying that MB (bytes) vs Mb (bits) isn't confusing? I talk to many, many people who sadly think Firewire is 400 MB/s and therefore can't understand the need for 800...

I don't think we need to resort to any new acronyms... also CCD vs CMOS doesn't fundamentally matter... all we need to describe is the sampling pattern.

This gives us:
3CCD - 3 monochrome CCD / CMOS sensors + a prism. These can use pixel shifting or not.
1CCD - 1 CCD / CMOS sensor with a color filter in front of it. This can be Bayer RGB, RGB stripe, Complementary color or RGBE (red, green, blue, cyan) so far. I can imagine future variations eg with ND filters on some pixels for HDR, etc... but let's just describe them as such.

In terms of placement of the sensors, you have...
standard - sensors placed on a standard grid
45 degree rotated - grid is rotated by 45 degrees, eg new Sony CMOS video cameras, Fuji SuperCCD
Fuji high dynamic range - a mix of large and small sensors for higher dynamic range

No need for acronyms, just string the words together - eg "3CCD pixel-shifted standard grid" and give the resolution of each sensor.

The only problem with the above that then 3CCD sensors are underrated - eg a 3CCD 2 megapixel sensor can definitely produce better images than a 2 megapixel single CCD sensor!

Of course, the Red site sadly appears to deliberately perpetuating that confusion!
On http://red.com/cameras.htm:
"Typical high-end HD camcorders have 2.1M pixel sensors"
should be
"Typical high-end HD camcorders have THREE 2.1M pixel sensors"

..especially if the next sentence says
"We deliver 12M pixels..."

Reality distortion field, anyone? Graeme? Jim?

I think the Red camera is good enough to crush its competitors on its own merits without help from un-professional Apple-esque mis-statements on its website!

If you change that line, I will be able to happily refer friends to red.com without giving them a disclaimer that "they lie about the specs a bit"...

Bruce

[Andrew M.]

Andrew... you're saying that MB (bytes) vs Mb (bits) isn't confusing? I talk to many, many people who sadly think Firewire is 400 MB/s and therefore can't understand the need for 800...


Agree 100%, I am just disappointed that some otherwise reputable companies resort to put just 3.3Mpixel (CMOS Bayer) next to referring to 2.2MP (RGB) so without adding what kind of sensor we have inside an average person could get confused.
Andrew