Digital Cinema Dynamic Range

This post opens a little window into my current thoughts about digital cinema, dynamic range, and some recent and ongoing testing of the RED One camera. I thought some of you might be interested in the process I go through when pondering things like this, rather than just some dry results.

Camera tests are a bit funny to me. You test a 3rd grader to give him a grade. You test the waters to see if you want to jump in. And you test house paint colors by pinning some swatches to the wall and trying to get a feel for what you like. I often feel that camera tests are more like that last example than anything else. But there is a kind of testing that makes a ton of sense—the testing you do when you've already 99% decided what camera you want to use, and now you want to figure out how you want to use it to produce a specific result. That's what writer/director Moses Ma did for his film The Ethical Slut. He and DP Paul Nordin knew they wanted to shoot with the RED One, so they staged a test based on a short scene from the film.

To make things interesting, the shot in an airy studio with big windows that they knew they couldn't control. And as you can see, those windows are overexposed, and clipped.

We all knew this was going to cause problems, but the exact nature of these problems has got me thinking about some of my favorite digital cinema questions: What exactly is dynamic range? And what constitutes a film-like image?

And of course a very concrete question: What should we do the next time we're shooting RED and there's a window in the room?

To help with my head-scratching about all this, I turned to After Effects, as I tend to do. I created a virtual scene with common exposure targets in HDR. Here's what it looks like:

And here are the targets annotated:

(Click on the image to enlarge it)

The gray card is your standard 18% reflectance, and it like all the cards in the scene are rendered without any shading so that they return pure values. The image, being a gamma 1.0 HDR, contains floating-point values that map 1:1 with the diffuse reflectance of the subjects, so the 18% gray card maps to pixel values of 0.18.

The black card is not pure black, but rather a more physically realistic 1% reflectance.

The pure white card is a bit of a hypothetical (nothing reflects 100% of the light that hits it), but it and the either cards are more about exposure targets than simulating a physical object. So we have spots in the frame that represent "white" and one, two, three and four stops over "white." We also have something that doesn't really exist in nature: pure black.

The "skin tones" in the scene are designed so that they both land on the "skin vector" of a vectorscope and rate as 70% luminance in Rec709. So between the skin target and the 18% gray card this scene is designed to be easy-as-pie to expose according to traditional rules-of-thumb.

Video rules of thumb, that is.

Here's the same scene underexposed by four stops so that you can see the relationships of all the exposure targets. This also makes it clear that each card has a patch of "detail" to help distinguish whether the truly falls within exposure.

All these images have been converted to sRGB space for viewing on your gamma 2.2 display of course. If you'd like to download the scene as a 16-bit float EXR, you can do so here.

So now we have our hypothetical scene, it's time to shoot it with a hypothetical camera. A Rec709 camera uses an encoding curve of roughly gamma 1.9, but its images are designed to be displayed at gamma 2.2, which is very close to sRGB, the viewing assumption for all images on this page.

So let's take a look at what happens when we shoot our scene with a Rec709 camera. Following one of many rules-of-thumb on the subject, we'll expose the gray card at 45% luminance and skin tones at 70%. Here are the results:

The same shot on a Waveform Monitor:

As you can see, the gray card is at 45% IRE and the skin tones are at 70%. But you can also see that the other cards are all pegged to the ceiling—even the 100% card. Exposing a pure Rec709 image using the common standards doesn't even leave you enough headroom for something "white."

If I expose the scene down by half a stop you can see the white card come into range in both the image and the scope:

So this is what happened on our RED shoot. We exposed according to video rules using the only image our RED One was capable of displaying, and we wound up with very little room for highlights.

You might ask what the big deal is here. "Sure, the windows (or cards) are blown out, but that's the picture we're making! The windows are bright and I want them to look white." That would be a perfectly reasonable position to take—unless you planned on transferring to film.

See, color negative film has a huge shoulder, the part of the famous s-curve at the top where it slopes gently off. "White" is nowhere near overexposed on film—it's right at the beginning of this shoulder, right after the straightline portion of the curve.

Watch what happens when I take my Rec709 image and convert it to Cineon Log using a gamma of 2.2:

Here I used the default settings of 10-bit black = 95, white = 685. As you can see, the white card and all its overexposed companions are maxed out at 685, nowhere near the maximum 10-bit value of 1023.

Here's what that image would look like printed to Kodak 2383 print film:

And now you see the problem. The clipped highlights map not to white, but to a pinkish gray that's about 80% of the available brightness.

You might suggest to change the white-point mappings so that those clipped whites get placed closer to 1023, but in doing so you brighten up the entire scene:

...but now our carefully metered gray card and person are way overexposed.

And this is exactly what happened with our test footage. We converted it to log using Red Alert and graded in under a 2383 preview LUT, and the windows appeared as pink, flat fields rather than sunny, overexposed sunlight:

Now the truth is, most Rec709 cameras save a little room above pure signal white for overexposure, and RED One is no exception. Using Red Alert I was able to extract a bit more out of those windows—but just a bit.

What traditional video cameras tend to do with that headroom is build a little rolloff into the Rec709 curve. Like a tiny version of film's mighty shoulder, this rolloff allows a more graceful clip into white whil still keeping things like gray cards on their waveform targets. Here's a hypothetical Rec709 image with a bit of a rolloff:

As you can see, gray is still at 45%, but now "white" maps to about 94%, and even the one-stop-over card is not fully blown out. This more closely approximates the behavior of most HD video cameras, and it would seem to be a good thing. What a clever person can do when transfering this kind of video to film is undo that shoulder curve and pull that extra bit of highlight latitude back up into the brighter Cineon registers without affecting the midtones, resulting in less flattened whites and as-expected mids.

But this concept, while critical for transferring old-school video to film, has nothing to do with RED, or any camera that shoots raw. Because how that rolloff works inside the camera is by underexposing, and then creating a linear image that is then bent into this funky bastardization of the Rec709 curve. So the camera that made the rolloff image above would first capture this image internally:

(note that the image has been converted to Rec709 for viewing)

...and then pushes up the mids to create this:

This is important when your camera records compressed, limited-bit-depth images. On Red it wouldn't help at all, because the gamma curve is merely a preview. But that doesn't mean we can't learn from the process.

The camera gave itself extra room for highlights by under exposing. A 1 1/4 stop underexposure gave the camera enough highlight latitude to create a nice soft shoulder.

And so we come to the conclusion that was, of course, blatantly obvious at the beginning: To get more highlight latitude, we'll need to underexpose. Obvious indeed, but what's maybe not so obvious is that without the rolloff tricks that we're accustomed to from our video cameras to fall back on, we can really screw ourselves by failing to underexpose RED One.

But how much should we underexpose? For one suggestion of how much, let's look at another digital cinema camera, the Panavision Genesis. The Genesis shoots to a color space known as "Panalog." Rather than using a gamma curve, it uses a logarithmic curve similar enough to the Cineon log curve that one can actually recreate the Panalog transfer function using a standard Cineon lin/log tool.

Here is our scene in Panalog, exposed per Panvasion's guidelines:

Note that while we still get to give our gray card a decent exposure, it's not up at 45%. And we have no trouble holding onto detail in the two-stops-over card. Lo and behold, this Panalog image transfers to film very nicely indeed:

(boosted 2 stops in printer lights)

Nothing magical here, although I do think there is real brilliance behind the Panalog format, since it can both be legible on a video display and transfer to film elegantly. But all that's happening is that the Genesis is recording a linear image on its chip that spans scene values from 0% to 600%. The log encoding is only mandated by the bit-depth of the recording medium. Armed with that knowledge, we can create a Rec709 version of the 0.0–6.0 linear image that the Genesis records:

And here we have the key. If we want the RED One to match the Genesis's latitude, this must be our exposure target in Rec709 (which, at the time of this writing, is the only monitoring option for the RED One, although that should change soon). We must ignore RED One's suggestions to expose it like a video camera.

Put 18% gray at 12.8% IRE.

Put white folk at 23.1% IRE.

"White" will land at 39.2% IRE, and you'll hold values all the way up to (and slightly beyond) 600%.

No problem. Except for two things:

This dark image won't be any fun to look at. Not for directors and not for DPs. Something more like Panalog would be better, and something to which a 3D LUT could be applied would be ideal.

But more importantly, RED One, like many cameras, has compression and noise, two things that show up like crazy in underexposed images. By standardizing on an underexposed image, we are putting the most important value ranges down in the mud where they just might get stepped on.

But hey, that's what Panavision's doing! They just happened to plan their entire camera and imaging pipeline around what they perceived as a digital cinema mandate: hold onto enough highlight values that your images look terrific on film.

So how will the RED One handle being underexposed by 2.580 stops across the board? We'll find out next week when Paul shoots his next round of tests!

Oh, and one last thing, just to bring this all home. We've seen how a Rec709 camera can't even hold 100% scene illumination without some underexposure, usually found in the form of a rolloff curve that allows it to capture maybe up to about 150%. We've seen how the Genesis cleverly holds onto 600% scene illumination. But what of film? Well, there are a lot of different film stocks out there, but as a rule the Cineon log model, which makes no accounting for toe or shoulder, maps Cineon log 1023 to 1,352%.

That's right, 1,352%. Here's what that looks like:

Only the 4-stops-over patch is clipped, and just barely (and it wouldn't be if the shoulder was modeled). Shoot that out to film and you get this:

A full range of exposure, rich skin tones, and subtle detail in bright highlights.

Just a little reminder of what we're all shooting for here.

Is RED One Really a 4K Camera?

Anyone interested in the RED One camera should read this comparative review/diary of the Sigma SD14 and the Canon 5D (the latter of which I am a delighted owner).

Both are SLRs, both are a few years old. But the 5D has a 12.7 megapixel, full-frame chip, where the Sigma produces a mere 4.6 megapixels. Why bother even comparing them?

The reason is that the Sigma has a Foveon sensor rather than a CCD or CMOS. This sensor can capture distinct R, G and B light information at every pixel. The 5D's CMOS chip can record luminance at every pixel, and uses the common Bayer pattern of color filters at the photosites to capture RGB color, intermingling color fidelity and spatial resolution in a way that must be decoded by software using some math, some compromises, and some guesswork.

One way of looking at this is that the 5D spends three of its 12.7 million pixels to accomplish what the Sigma achieves in only one. If the 5D records 12.7 million tiny little light records per image, then the Sigma records 13.8 million (4.6 x the unique R, G and B records per pixel).

But that's not entirely fair. Our eyes tend to perceive detail more in luminance than in color, and the 5D is recording much more luminance information that the Sigma. For black and white photography, the 5D truly is a 12.7 megapixel camera and the SD14 truly is a 4.6 megapixel shooter.

So the truth lies somewhere in between—the SD14 is neither the 5D's equal in resolution nor is it possessed of one third the pixel count.

In one very real way the RED One is a 4K camera. It creates 4K images that look damn good.

And in another equally real way, the RED One is not a "true" 4K camera, as each of the 4K's worth of pixels it creates for each frame is interpolated (from compressed Bayer data at that).

And that's probably just fine—a topic for another day.

Don't Panic

So hey, that RED One camera is in the hands of its first customers, and there has been an explosion of traffic, images, enthusiasm and confusion about it.

Mostly confusion.

Shooting raw is a new thing for a lot of people. Even those of us who are used to shooting raw with DSLRs aren't accustomed to seeing our images with little or no post-processing—when you open a raw file in, say, Lightroom or Aperture, the software tries to make the image "look good" for you (non-destructively, of course), and lets you tweak from there. But RED Alert, the beta software that ships with RED One, doesn't do this. Nor should it, but it RED Alert has enough controls to daunt many new RED One owners.

In truth, RED Alert is probably hurting matters by offering too much control too prominently in its UI. Most RED One shooters would be better off setting white balance and nothing else (consistently per sequence), and then selecting between a hard-coded preset for either Rec709 gamma (for video post) or log (for film-style DI post). To do any more tweaking than that in Red Alert is to simply muddy the waters and cause downstream confusion.

Many people, for example, feel the need to correct for underexposure in Red Alert. I've seen people apologizing for underexposed test shots. Don't—underexposing is exactly what you should do, within reason, in order to hold highlight detail. If you look at the "offhollywood" test shots on hdforindies.com, you'll see that the exposure is all over the map. That's fine! That's a very easy thing to correct for in post, and holding onto those troublesome highlights is worth some inconsistencies from shot to shot. Remember, the dynamic range of a digital camera has nothing to do with how much overexposure it can handle (because no digital camera can handle any)—it's all about how much you can underexpose. In other words, as you try to hold onto that highlight detail, how much can you underexpose that car before it reveals nasty noise, or worse, static-pattern artifacting, when you brighten it back up in the DI?

I only wish that Mike and company had transfered every single one of those shots with the exact same RED Alert settings—it would be so much more illuminating.

Graeme Nattress, author of RED Alert and chief image nerd at RED, started a thread on RedUser.net in an attempt to guide people's initial use of RED Alert. I joined in and added this:

I would appreciate it if people posting example images would differentiate between attempts to simple "develop" the RED One image into a workable video form vs. attempts to make the image "look good."

The reason being that some people will be looking at these images for proof of RED One's empirical qualities, i.e. dynamic range, highlight handling, ability to hold detail in saturated colors, etc. These people will be disappointed to see a clippy, crushy image that has lots of sex appeal and "looks good."

And of course, some people will be looking at the first RED One images off the line and hoping that they "look good." But that should not be the case unless the images have been color corrected. While RED Alert has some color correction controls, it's not a color grading station, and the ideal RED One workflow would most certainly not be to make permanent color decisions early in the process.

Remember that an image that shows a broad dynamic range will look flat and low-contrast. An image that shows good highlight handling will probably appear underexposed. And an image that shows good color fidelity will appear to have very low color saturation! I urge new RED One users to learn to love underexposed, low-con, low-saturation images as they come off your camera, for they contain the broadest range of creative possibilities for you later.

But also maintain your love for the rich, saturated images that you may ultimately create from this raw material—and hope/beg/plead for tools to allow shooting with RED One under a non-destructive LUT that is included with the footage as metadata, so that you can preview your image as it may ultimately appear, record that nice flat raw image, and later have the choice of applying your shooting LUT or some other awesome color correction.

And then I went and listened to the fxguide podcast about Mike and Jeff's first day with the camera, and felt tangible pain as these incredibly sharp guys verbalized their near-terror at the learning curve that lays before them. Guys, it's so much easier than you think. Don't stress out about what to do with RED Alert—the less you do, the better. And so much more the better if you do the exactly same thing to every shot.

Next time: What do do with all these flat, low-con, underexposed and uneven—but consistently processed—images! The good news? If you've read The Guide, you already have a leg up.

Should you buy a Red One?

It's no secret that the Red One has caused quite a stir, and for good reason. The price point, design philosophy, and Howard Hughes-ian nature of the endeavor are thrilling and a welcome change in the staid world of camera manufacturers. I'm a big fan.

And money has now been placed where previously did mouths only tread. We saw working cameras at NAB, and some truly amazing footage shot in real-world conditions. So the Red One is real, and you could probably buy one if you wanted to. Do you want to?

I have a reservation for a Red One camera. Or rather, The Orphanage does. There's no way I'd be putting my own $17,500 into a camera body, no matter how badass it is. But for The Orphanage, it almost makes sense. With the body, a Nikon f-mount front plate, and lenses we already have, we can use it for element shoots and spec projects. Would we do enough of that in a year or three to make it cost-effective to buy rather than rent? Maybe. Who knows, when our reservation comes up, we may not even buy the thing.

Sometimes I get reminded why I don't post more on public forums. One such occasion came when I dared to gently point out on DVXuser.com (a forum about accessible filmmaking, one of my favorite subjects) that a satisfactory filmmaking kit centered around the Red One would cost a great deal more that 20-or-so grand. I actually received personal messaged from people who were incensed that I would dare try to chip at their dream, accusing me of propagating false numbers.

There are those out there who are throwing rocks at Red because they feel threatened by it, or because they like the status quo of cameras made by and for the elite. So I have to be careful in my writing lest I be mistaken for one of those people. In fact I'm the biggest fan of accessibility there is. I dare say I wrote the book on it. I'm just concerned that some Red reservation holders are getting themselves into the same situation I did when I was about eight years old, and I saved my allowance for weeks to buy an AT-AT toy. I saved and saved and I even factored in sales tax, so that when I got to the store I had exactly enough change in my jar to buy the Imperial Walker.

But not the batteries.

Luckily my Mom was there to bail me out. But she won't be there for me when I realize that my razor-sharp Nikon 50mm prime (which mounts to the $500 Red f-mount plate) breathes like crazy when I try to rack focus with it (Red 18-50 f2.8 zoom: $6,500), or that I really quite need a follow-focus regardless of which lens I use (cheapest follow focus I've found: $645, rails to mount it to: $1,250), or that a tripod that can hold a fully kitted-out Red One costs a couple grand at the very least, or that I might want some way of seeing what my camera is shooting (Red EVF: $2,950) or some way of recording it (Red Drive 320GB: $900). Mom?

Could you strip down your expectations and start making pictures with a Red One for about $20,000? Absolutely. But should you? Will you make a better movie with this minimalist Red setup (one lens, no tripod, 720p LCD monitor) than you would with an HV20, an M2 adapter, that same Nikon lens, and $17,000 left over for things like extra lenses, DV tapes, and coffee for the crew? Which of these scenarios creates a better experience for the audience? Which puts you, the filmmaker, in a better place to succeed?

Although it may sound like those questions are purely rhetorical, they are not. As you will recall, I have a reservation and I must decide what to do with it. These are real, active questions for me, and, I imagine, for many other people.

So in a world full of entirely justified optimism and excitement about the Red One, I hope you all will indulge my exploration of the elephant in the room. No matter how bitchin' the Red One may be, it may not be the best camera for you. Or me.

I'm going to finish with a lengthy quote from Carlos Acosta, a DIT from LA who posted this eloquent reply to a conversation on the CML about the relative merits of buying any camera.

...Let me throw some great reasons to NOT own production hardware. They will be especially true if you are a post/consultant. I am amazed how many producers, editors, animators, etc have committed to buying a camera.

1. Production gear gets beat up and blamed for every problem that comes afterwards.
2. No matter how cheap you think it is, it's expensive.
3. You will never have the right lenses
4. You will never have the right tripod.
5. You will never be cheap enough.
6. It will break in the middle of a shoot and you become public enemy #1.
7. You will find yourself changing your career so that you can keep your fancy camera busy even when it is the wrong camera for the job.

This list could go on and on....my only hope is that those who buy production cameras and related gear understand the TOTAL COST OF OWNERSHIP. It's not as simple as adding up the hardware. If you buy equipment for the purpose of making money, it's a business. If your current business is not directly related to renting cameras you will have to start a new business which will likely compromise your old business. Now, your new camera business has to have insurance, clients [the bigger ones like free lunches], an office, subrentals, repairs, new gadgets, LOTS of time on the phone explaining the new technologies to a clueless producer hoping for a cheap camera.

All of a sudden....$17,500 seems scary to me. When it's all said and done, it's more like $90,000 plus continuing operating expenses and I still have an experimental system that has been branded as the CHEAP alternative. It's still a camera, the business has not changed at all. For the fifteen years I've been in this business, I have seen countless products that promise the world and actually deliver. Did the world change? No, it just got more pixels, more color, more speed, or more sex appeal.

I purchased a DVW-700 in 1997. It was 1035i and the hottest and most exciting development for digital acquisition at the time. It was quickly changed to the HDW-700a which did 1080i. It was a BetaSP/DigiBeta world at the time and no one knew what to do with this camera that required huge changes to take full advantage of. I had the newest toy on the block and was the poorest guy around for it.

The thread about compression is a great example of being realistic in a business that is full of crazy expectations. Compression is a reality for a multitude of reasons, but it comes down to making money. Buying a camera and what camera to buy requires a glimpse of reality as well.

If, on the other hand, you want an awesome toy and have lots of excess cash.....

Carlos Acosta,
DIT/Engineer/Reality Checker
LA

Thanks for permission to re-use that here Carlos. The conversation is still active there on CML, here, and everywhere else. Is the Red One the most exciting thing to happen in Cinema in years? Heck yeah. Does that mean you should buy one? Maybe, or maybe not.

More to come on this, of that you can be sure.

Red One camera body: $17,500. Red One as pictured above, $33,870 (my best guess based on prices at the Red Store)