Whites and blacks colo(u)RED by...

[Emanuel A Guedes]

http://www.reduser.net/forum/showthread.php?t=10456

[Chris Keth]

Can you repost the images here? I am not a member at reduser.

[David Mullen ASC]

Err, most HD cameras can handle whites in a soft-lit room as long as you set the exposure for where you want white things to start clipping or not start clipping.

 

The problem or challenge with dynamic range comes in more extreme situations than what those clips show -- like someone standing in the deep woods with a sunlit field in the background, where the background is several stops overexposed but if you exposed so that there was no clipping, then the shadowed subject is several stops underexposed.

 

But someone wearing white in a soft-lit interior, you just have to set the whites under 100 IRE if you want some texture in them, even if that forces the faces down a stop -- correcting in post for that is not hard since you have all the information you need there.

 

It's harder when you introduce, let's say, a slash of sunlight hitting the white areas in the room or wardrobe and you want the sunlight to be very hot, like three stops over, but you also don't want unnatural amounts of clipping.

 

I mean, the shots posted are lovely but I don't see them as a particularly worst-case scenario for dynamic range.

[Brian Drysdale]

The forest scenes in "Apocalypto" with the hot areas, were the ones in which the Genesis looked most like video. Practical lamp shades with wall/table highlights in low key scenes shot on HD are often not handled as smoothly as film.

 

The fire test in the Dalsa forum looks pretty smooth, although the object in the fire is close to the flames.

[Guest tylerhawes]

The fire test in the Dalsa forum looks pretty smooth, although the object in the fire is close to the flames.

 

FYI the film that test is from is called Wick and we are doing a 4K DI for it now. We're actually grading straight off the RAW files, which I believe we are the first to do, and it is a huge boon to the workflow to be able to do that. Besides the time savings, as a colorist I love having access to all the information from the camera without having to go back to an earlier software pass to get at it. It's also pretty impressive technically to see that systems are so fast now as to be able to grade 4K RAW in realtime.

[John Brawley]

It's also pretty impressive technically to see that systems are so fast now as to be able to grade 4K RAW in realtime.

 

Using which platform ? Viewing using ?

 

jb

[Guest tylerhawes]

Using which platform ? Viewing using ?

 

We're viewing it on 2K projection and can toggle between 2K fit and 1:1 zoom. In zoom, we can pan around the image to checkout details. Right now the 2K projectors have better colorimetry (or at least they did 6 months ago when I last compared, but I know Sony is improving all the time). For DI, I prefer the better color than to have all the 4K pixels, especially since I can always zoom in if I need to. You have to stand up and walk toward the screen to see all the 4K detail even if you are on 4K projection anyway, so the 1:1 zoom in the DI system seems just as valid to me.

 

As far as the system, I'm not sure if I can say yet, but I'll find out. I don't think it's exactly a "secret", but I should make sure before I blab too much about it. I can say that it's working of a single fast Xeon workstation with Quadro card, so we're not talking render farms in the backroom to make this happen...

[John Brawley]

We're viewing it on 2K projection and can toggle between 2K fit and 1:1 zoom. In zoom, we can pan around the image to checkout details. Right now the 2K projectors have better colorimetry

 

 

 

I don't mean to be difficult, but you gave the impression you were grading in 4K in real time. Except you monitor in 2K and play back in 2k.....:-)

 

I gather you are suggesting that you are grading in Scratch ? Having done some 2K and 4K DI's in scratch, it doesn't seem to me that you actually NEED to grade in 4K anyway.

 

jb

[Guest tylerhawes]

...you gave the impression you were grading in 4K in real time. Except you monitor in 2K and play back in 2k.....:-)

 

Yes, I'm grading in 4K RAW in realtime, your impression is correct. I play the 4K timeline, which the system scales down to 2K on output to the projector, just like virtually everyone else offering 4K DI. Having 4K projection is much less important than having the 4K RAW pipeline, and is almost another consideration entirely to me. What IS remarkable is that the system is working with 4K RAW files, debayering, grading primaries/secondaries, applying pan/scan, 3D LUTs, and outputting in realtime. That it has to scale to 2K on top of it is just all the more impressive, not a detraction.

 

I gather you are suggesting that you are grading in Scratch ? Having done some 2K and 4K DI's in scratch, it doesn't seem to me that you actually NEED to grade in 4K anyway.

 

I'm not using Scratch.

 

I think you're missing the point of why this is a big deal, because the fact is it is a huge benefit for this film:

• No intermediate step of converting from RAW files to 10-bit DPX (saves time and disk space)
  
• Working straight from RAW, I have access to the full "digital negative" in the grading system, without having made any destructive color decisions in the conversion process. It is a much fatter negative than 10-bit log
  
• I can make all my zooms, repos, pan/scans and see the final quality on the projector in realtime (if working with 2K proxies, I wouldn't know how well the zoom holds up until I switched out the proxies for the 4K plates)
  
• At any time I can hit a button and zoom 1:1, pan around the image and checkout the details on the 2K projector
  

So it's not just about the 4K resolution, it's also about having access to all the RAW color information. The benefit of both 4K and RAW is essentially over sampling. Over sampling on resolution, over sampling on color, so that I can make all the decisions I want and know there is plenty of headroom to work with, and see it interactively so I know how far I can push it.

 

Classic example of old vs. new way that is real-life from today:

 

A waist-up shot of woman looking out a window. The room is in shadows, which envelop the side of her face, while the front is lit softly by the window light. There is a shard of hard light that falls across the edge of her face and down over her hand, which is around her stomach.

 

That's a tough shot. Huge range. We want the shadows to be strong, but not crushed. Highlight should feel edgy, but textured. The fill on her face (filtered light from window) will need some curves to sit right. We also want to zoom in midway on the shot for an extreme close-up.

 

Now let's look at old way vs. new way as it applies to this shot:

OLD WAY (RAW files > debayered 10-bit log 4K DPX > 10-bit log 2K DPX proxy)

Right away you run into some contrast issues, because it is a real struggle to get the nuanced curves in the highlights and in the shadows without starting to breakup the image (banding/speckling). There's just not enough information to get a smooth contour to the contrast while achieving everything we want to do. Solution? Go back to the RAW files and reconvert this one shot, taking special care. Only problem is, the RAW conversion software is going to have a much more rudimentary set of color tools, and probably isn't setup to run in the DI suite, so there's going to be some trial-and-error, and you only have so much time... Then you try out the repo, working on your 2K proxies. Before the repo, it looks great of course because you have a 2K image on a 2K projector. But on this shot, we want to zoom WAY in, from a waist-up shot to tight on the face, giving her a haircut. Do this on the 2K, and it's looking very soft, wouldn't be able to pull it off. Of course we know the 4K will look better when we reconform, but we're guessing how good it will look until we swap out the proxies for the 4K DPX plates. How easy that will be and the interactivity you have tweaking the repo will depend on the system.

 

NEW WAY (RAW files native)

Same situation, so simple. I have my complete DI toolset at my disposal to tweak the contrast curves and get them just right, knowing that no information from the RAW digital negative has been wasted during the intermediate conversion (that I didn't need to do). So I quickly dial in the curve and have exactly what the DP wants for this nuanced shot. Now for the repo, I zoom in about 300% and, wow, if I didn't know better I'd think there was no zoom at all on this shot. Good thing I had the 4K pixels to see that for sure, and not guess about it, or at the very least not spend any time thinking about or switching to another set of files (if I'd been working on proxies). IOW, it's just a creative workflow, no techie workarounds, no extra steps. Just get your footage and squeeze what you want out of it.

BONUS: THE IT ADVANTAGE

Besides all this, there is the management angle, which is that instead of 16MB/frame RAW files AND 48MB/frame 4K files AND 12MB/frame 2K proxies (count 'em, that's a total of 76GB/frame of source files!), I only have the one set of 16MB/frame RAW files (a savings of 79%!). Plus, because the RAW files are only a little bigger than the 2K files, our existing storage (which was over-engineered a little bit anyway) handles them with relative ease.

Bottom line to you as the client: more creative, faster, higher quality, and cheaper than the RAW>DPX 2-step. What's not to like?

 

 

Edited March 25, 2008 by tylerhawes

[John Sprung]

NEW WAY (RAW files native)

This is interesting. You're working on a Dalsa show, so what exactly are its RAW files? Is it a Dalsa specific format, or is there an open specification? Could Red's R3D format be decompressed to the RAW format you're using? The big hurdle in any digital workflow is getting everything to talk the same language.

 

I especially like that everything the chip saw is in your hands, no need to go back to anything. That's definitely the right answer.

 

 

 

-- J.S.

[John Brawley]

What IS remarkable is that the system is working with 4K RAW files, debayering, grading primaries/secondaries, applying pan/scan, 3D LUTs, and outputting in realtime. That it has to scale to 2K on top of it is just all the more impressive, not a detraction.

[/b]

 

 

I was under the impression that Scratch did all of this as well. Are you inferring that you are beta testing another platform which performs in a similar way to scratch ?

 

jb

[Guest tylerhawes]

This is interesting. You're working on a Dalsa show, so what exactly are its RAW files? Is it a Dalsa specific format, or is there an open specification?

 

RAW files are proprietary to the model of camera they came from. The idea of a standard RAW format seems cool, but the whole point of RAW is to have everything from the sensor uncompromised, so it makes it a challenge. The best example of where things might go in a few years is looking at Adobe Photoshop and stills, where Adobe has established a standard framework that companies can use to have easier compatibility with Photoshop. Still, I had to upgrade to CS3 to get it to work with my still camera, so it's not a silver bullet.

 

But for post right now, every camera has its own RAW format. Dalsa sends data that gets recorded as "DPX" 4K files. If you open them up, they are grayscale and have weird digital patterns in them. They are not really pictures, they are data. So they are DPXes, but proprietary.

 

To use them they have to be interpreted, and that is where the power of grading from RAW is, because you do the interpretation in your DI system. Imagine if you had the ability to develop film using all the tools of a DI system working right with the unexposed negative, think of all the control and detail you could get out of it. It's sorta like that.

 

Could Red's R3D format be decompressed to the RAW format you're using?

 

 

.R3D files are already RAW, so you'd want to work straight from them, not convert them. Scratch does this now. The system I'm working in can also do it with some older .R3D files I have, but the latest ones are encrypted so it won't be until RED comes out with the SDK that this particular system will grade RAW. In the meantime, I'd probably use Scratch for RED films (I used it for many months on Superman Returns so feel totally comfortable switching back and forth between systems).

 

One difference is that .R3D files are 12-bit wavelet compressed RAW files, while Dalsa's files are 12-bit uncompressed RAW files. The RED compression is very good so I'm not knocking it, just something worth knowing. The nice thing about compression is that it saves space. The flip side is, well, it's compressed :)

[Guest tylerhawes]

I was under the impression that Scratch did all of this as well. Are you inferring that you are beta testing another platform which performs in a similar way to scratch ?

 

AFAIK, SCRATCH does this with RED .R3D files, but not any other camera RAW files. I suppose the work they've done figuring out support for RED camera RAW will help pave the way for supporting other formats in the future. I think all DI systems will be forced by the market to support RAW for any popular camera systems over the next year or two at the most, but right now there are a few early leaders.

 

The one other note is I think Scratch is limited to doing a medium/high 2K extraction of the 4K .R3D files, not 4K. Assuming you have very fast storage and that is not a bottleneck, .R3D files are actually harder to debayer in realtime because they are wavelet compressed. That means before the DI system can debayer, it has to uncompress the files also, so it's more work. Anyone who has ever worked with JPEG2000 4K files knows what I'm talking about: serious horsepower.

 

Apparently an overclocked, liquid-cooled Xeon system can get Scratch doing 4K .R3D, or close to it, so it seems like all we need is to give Moore's law a year, give/take, and then maybe that can be done...

 

I don't know if Lucas haunts these parts, but he can correct me if I've mispoke on anything...

[Lance Flores]

I was under the impression that Scratch did all of this as well. Are you inferring that you are beta testing another platform which performs in a similar way to scratch ?

 

jb

 

We tested four of the coloring and conforming post products and chose the Piranha Cinema 5. It did much more for far less the cost and had the fastest render of all the machines. We're purchasing two that will have multi fiber channels connections to the Linux SAN with all fiber channel drives. This was the fastest, and most reliable system we found that had everything we wanted and interfaced well to our Final Cut Pro, sound systems, in a 4K pipeline. If anyone interested in seeing the set up we'll be four-walling the system over at Grace & Wild, Inc. studios, Farmington Hills, Michigan in a few weeks.

[Guest tylerhawes]

We tested four of the coloring and conforming post products and chose the Piranha Cinema 5. It did much more for far less the cost and had the fastest render of all the machines. We're purchasing two that will have multi fiber channels connections to the Linux SAN with all fiber channel drives. This was the fastest, and most reliable system we found that had everything we wanted and interfaced well to our Final Cut Pro, sound systems, in a 4K pipeline. If anyone interested in seeing the set up we'll be four-walling the system over at Grace & Wild, Inc. studios, Farmington Hills, Michigan in a few weeks.

 

I think Piranha is maybe the only system I haven't laid hands on. Once in a blue moon I hear about Piranha and then I forget about it until the next time. They have almost no marketing presence that I'm aware of. If they have a good system, they should get out there more and tell people about it.

[John Sprung]

RAW files are proprietary to the model of camera they came from. The idea of a standard RAW format seems cool, but the whole point of RAW is to have everything from the sensor uncompromised, so it makes it a challenge.

 

I think the way to handle it would be to standardize a wrapper format. Put all the necessary parameters to understand the data in a standardized header, followed by the data. Raw from a Bayer or striped sensor would have to specify how many photosites wide by high, which colors are on which photosites, bit depth, log/lin, etc. That's not complete, of course. It's something for a standards committee to hammer out.

 

Dalsa sends data that gets recorded as "DPX" 4K files. If you open them up, they are grayscale and have weird digital patterns in them. They are not really pictures, they are data. So they are DPXes, but proprietary.

 

Is it all nonsense, or does it look like a mildly solarized B&W image? If it looks like an image, it may be presenting the data from the photosites without any color, as if it were a B&W sensor. With that, you could make a viewable image just by putting the appropriate colors on the pixels, and letting the eye/brain do the de-Bayering. That would work for proxy/thumbnail quality, of course you'd need to do real de-Bayering for the final. But that kind of thing is how film works -- grains are on one of three layers, each grain get only the color of its layer, and our eye/brain combination does the rest.

 

One difference is that .R3D files are 12-bit wavelet compressed RAW files, while Dalsa's files are 12-bit uncompressed RAW files.

 

What we'd have to do with them is just the wavelet decompression, the output from that could be put in the RAW wrapper/header system.

 

 

Thanks for all the info, these reports from the cutting edge are very valuable.

 

 

 

 

-- J.S.

[Guest tylerhawes]

Is it all nonsense, or does it look like a mildly solarized B&W image?

 

The latter.

[DJ Joofa]

The one other note is I think Scratch is limited to doing a medium/high 2K extraction of the 4K .R3D files, not 4K. Assuming you have very fast storage and that is not a bottleneck, .R3D files are actually harder to debayer in realtime because they are wavelet compressed. That means before the DI system can debayer, it has to uncompress the files also, so it's more work. Anyone who has ever worked with JPEG2000 4K files knows what I'm talking about: serious horsepower.

 

My understanding is that R3D files are wavelet compressed, therefore, it is easy to extract a 2k image from the 4K wavelet co-efficients without doing the full decompression. May be that is the reason scratch does 2k out of 4k.

 

The point is from wavelet data one can have faster smaller size preview (pretty good quality) for color grading reasons without doing full decode and hence saving time/horse power.

[John Sprung]

The latter.

Thanks, Tyler. It would be fun to look at some of those images in uncompressed .BMP files. I'm sure that the DCT compression in JPEG would pretty much scramble the color differences.

 

 

 

-- J.S.

[Illya Friedman]

Dalsa's files are 12-bit uncompressed RAW files.

 

I just discovered this thread. One small correction- DALSA RAW files are actually uncompressed 16-bit .DPX.

 

I.

 

Illya Friedman

DALSA

Digital Cinema Division

www.dalsa.com/dc

[Illya Friedman]

Thanks for all the info, these reports from the cutting edge are very valuable.

 

Agreed!

 

I.

 

Illya Friedman

DALSA

Digital Cinema Division

www.dalsa.com/dc