How the Alexa really works

[Carl Looper]

Carl, I am never one to advocate 'dumbing down', but is there any way you could write a post that doesn't need a thesaurus and a dictionary to understand?

 

Ha ha.

 

Which terms require elaboration?

[Tom Yanowitz]

 

It depends on whether you see shadows as an indispensable part of lighting, and the thing that gives light shape, or you see them as a math problem.

 

https://www.youtube.com/watch?v=zSZNsIFID28

 

Edited March 10, 2016 by Tom Yanowitz

[Stuart Brereton]

The difficulty with regarding dynamic range purely as a math problem is that it demands a mathematical solution. You've managed, with your charts and figures and values, to illustrate that the solution is to shoot at 160ISO, which will give you the most even spread of values between highlight and shadow.

 

Unfortunately, cinematography isn't just a science, or even mostly a science. There are numerous real world factors that significantly impact the way we work. Constraints of time and equipment force us into compromises every day, and not just in how we choose to expose our images. Every project's ambitions outstrip its budget, and DPs are required to do more and more with less. Tripling my lighting package so that I can shoot night exteriors at 160 ISO is simply not going to happen.

 

So, we compromise. We trade off some noise for some speed. We lose some shadow detail in return for the ability to shoot within our means. That compromise might be mathematically 'wrong', but it's 'right' in every other way.

[Tom Yanowitz]

Yeah sure, that's why I rarely work at 160, for the same reasons (student films, we're glad we even have the chance to shoot with the Alexa from time to time, but the light sources ain't no 18K)

 

Now you might fairly ask why all the fuss of this thread then.

- Well first to be clear I didn't started this topic to ramble about ISO, It wasn't my point initially but somehow every one started talking about it.

 

- And to limit the use of high ISO at when you can't do otherwise in practice, even if that happens to be 80% of the time.

 

For example I was convinced earlier in my studies that I should shoot day exteriors at 1280 for highlights range (I know right).

But the exercise that day was to shoot at widest aperture (1,4) cause it was some kind of exercise on cinematic blur or whatever.

Now imagine the amount of stacked NDs in front of the lens, that was ridiculous, just because I wanted 8,5 stops between mid gray and sat (maybe I wanted to expose both for the back-lit actor and have some nice details on the sun itself :rolleyes: )

Result : Needlessly underexposed and noisy footage by the damn amateur that I was, and of course the highlight range not even close to used.

 

(By the way Deakins has been taken by the 'high ISO' political group on this thread, but he does shoot day. ext at 400.)

[Stuart Brereton]

I think you'd find most DPs try to shoot at a lower ISO when they can. I vary all the time between 400 - 800, and occasionally go as low as 250 or as high as 1600. In the very beginning of digital cinema, with Digital Betacam and then HDCam, it was common practice to shoot at -3db gain for a cleaner signal, so it's not an issue that people are unaware of.

[Satsuki Murashige]

Also, part of the problem is that as a student you often don't have access to the kinds of lighting and grip equipment or the crew to properly use them that would allow you to get optimal results. I know that I didn't. And when you do, you're still learning how to use them. It takes awhile to realize that the camera format matters a lot less than lighting and grip when it comes to producing polished work. When you start working for DPs and gaffers who deeply understand lighting and exposure, it is an eye-opening experience.

[Carl Looper]

 

https://www.youtube.com/watch?v=zSZNsIFID28

 

 

Richard Feynman is great. His work in quantum physics is awesome.

 

I'm not sure about what is being argued in the second video. Analogy is a really important aspect of film making. What is a photographic image if not (amongst many things) an analogy? The interviewee in the video is assuming that the source of any analogy is that which is already given. But one of the most important things in cinematography is actually creating that which will be the source of the analogy. What is proactive lighting if not altering the given. In many ways the result is always new in the sense it's always unique (can't be repeated in quite the same way). Unless of course one is shooting something like a test chart.

 

Secondly, first principles don't guarantee any external assumptions will be correct. But before getting to that, first principles are definitely a good idea. For example, they allow one to re-derive something from scratch, where one might have otherwise misplaced a rule of thumb, or a more complex formula. I'm often going back to first principles simply because I can't find some particular book in which a derivation was otherwise documented. But I'd certainly prefer if the book wasn't misplaced. Would be a lot faster. But it's also the case that a previously derived formula may not be quite relevant to a given task. One may have to go back to first principles in order to derive a new branch more specific to the task at hand.

 

While mathematics is incredibly useful it's almost invariably incapable of being a full solution, because in it's practical application, it works with assumptions outside of mathematics - and those assumptions can easily be incorrect. For example, one might design some setup in which various factors have been mathematically solved and realtime software written accordingly, but some "idiot" responsible for building the setup decides it won't matter if they tweak the design a little. He's run out of wood and makes a workaround, and doesn't tell anyone. Without attention to such possibilities a mathematical solution can end up idealising otherwise incorrect assumptions in play (such as a builder knows the purpose of what he's building). The math itself won't be the issue. In itself it's never wrong (we think). I mean, if you make a mistake when doing mathematics it's not the mathematics which will be in error. It will be because you will have done something that is mathematically incorrect (such as 2+2=5). Or otherwise made an incorrect assumption (turning up for an eclipse on the wrong day).

 

One eventually learns to how take into account more than just the maths. To predict all the vagaries that can occur external to the maths, and factor such into a solution. To make a thing in such a way that it can be recalibrated on location to account for all the conceivable vagaries that could occur. Statistics becomes more important than mathematics in this regard. And in some cases, particularly quantum physics, there is no mathematical solution (it would seem) - there is only the statistical solution. Something Feynman would definitely understand.

 

And thirdly, of course, mathematical solutions presuppose a problem, whereas a lot of creativity is about creating an interesting problem to solve in the first place.

 

C

[Brian Drysdale]

This area seems to be less about mathmatics than test, test test. There are lots of variables, including accidents of chance that'll either work to your advantage or against you. If you check out CML, you'll find loads of tests, with models, in of front windows, black backgrounds and test charts. Thus allowing things to start becoming interesting when you start breaking rules. Also, restrictions can force you to become more creative.

[Carl Looper]

This area seems to be less about mathmatics than test, test test. There are lots of variables, including accidents of chance that'll either work to your advantage or against you. If you check out CML, you'll find loads of tests, with models, in of front windows, black backgrounds and test charts. Thus allowing things to start becoming interesting when you start breaking rules. Also, restrictions can force you to become more creative.

 

Yes, testing is good. In the area I work, I write software and this gets sent to a tester who has absolutely no idea of how the software works - which is actually a good thing, because they are free of any assumptions I've otherwise built into the software. They'll bring their own assumptions to the software and in doing so are able to identify anything in the software that clashes with my assumptions.

 

And any clash is a good idea to address, because it means that same clash could occur elsewhere, ie. in other users of the software.

 

As a programmer, I know exactly what I'm assuming because I will have formalised it all in black and white. It's there in the code, staring me right in the face. The tester has no such access. They don't know how to read the code. But they know how to run the code and test it. And they can come to me with anything they find differs from what they were expecting. And we can, together, change it - if required.

 

But some testers think it's only their preconceptions which matter - and what's worse, that anything different from such must be some sort of incompetence on behalf of the programmer. Even though they couldn't write a program if their life depended on it.

 

Some testers also don't quite understand that what you write determines exactly what the result will be. A producer with whom I was working once (who came from a backgound as a software tester) asked me to test some code I'd written (as we didn't have a tester) and I said to him - look you do it - I already know it works - you don't. After a lot of shouting and me walking out he eventually sat down and tested it - and unable to find any fault in it, spent the rest of the day playing it over and over again, convinced there must be some fault in it. But the only reason there was no fault in it is because we both had the same expectation of what the software should do. The difference is that I could implement that expectation whereas he couldn't. And no amount of testing was ever going to change that happy state of affairs. But he spent all day trying. I felt very sorry for him. Not that the prick would ever understand that.

 

C

[Brian Drysdale]

Like other users, I do beta testing on the Lightworks NLE and it's not unusual for new sets of issues to a arise once a release version comes out because it has a wide range of users from feature film editors to people editing their games for YouTube. This always seems to throw up the unexpected because many users don't have preconceptions (or even industry standard workflows) and are workimg on a wide range of computers, many of which well below the recommended specification. It adds a tension to keeping the unique aspects to a piece of software and having a program that's the same as every other NLE, so losing it's selling point.

[Carl Looper]

Like other users, I do beta testing on the Lightworks NLE and it's not unusual for new sets of issues to a arise once a release version comes out because it has a wide range of users from feature film editors to people editing their games for YouTube. This always seems to throw up the unexpected because many users don't have preconceptions (or even industry standard workflows) and are workimg on a wide range of computers, many of which well below the recommended specification. It adds a tension to keeping the unique aspects to a piece of software and having a program that's the same as every other NLE, so losing it's selling point.

 

Oh certainly. One is dealing with a whole range of expectations. Sometimes, of course, there is an obvious mistake, ie. it's obviously meant to work this way rather than that way, eg. the computer crashing when you click on a particular button is obviously not going to be something anyone wants. Such problems are really easy to solve because every ones on the same page (ie. nobody wants it). All that's required is a fix. It's everything else that is a more difficult problem. What features should the software have that it doesn't yet have? How should it behave? Anywhere there is a difference of opinion one has to weigh up what to do. In short, it's not a computational problem. It's a "political" problem. Once that's solved the rest is just computational.

 

C

Edited March 12, 2016 by Carl Looper

[Noah Yuan-Vogel]

So Arri is only relying on the lower gain ADC for the top two stops of DR to overcome the DR-limited ADC? No other special stuff going on? So in a way the combined ADC is effectively 14.8bit I guess?

[Tom Yanowitz]

I guess

( log2(2*16384-4096) was your calculus right ? )

 

It just means that if you were to compare the Alexa 16bit raw (which you can't because you don't have access to it anyway) and a legit 16 bit sensor (doesn't the Sony F65 have one ? not sure) :

on the Alexa the two brighest stops would only have one value used every 4 values (analogous to identical pixels for a dvd upscaled to a bluray), against one for one on a 16bit adc.

 

But it's so much overkill highlight information anyway that it doesn't really matters I think.

Edited March 19, 2016 by Tom Yanowitz

[Sebastian Ziegler]

What happens when you shoot ISO160 under expose to not clip your highlights and then bring it back up to 800ISO in post? Does the give you the same results as shooting at 800 in camera or are the results worse?

[Tom Yanowitz]

For Arriraw : same results.

So basically if someone wants to look at the image on set, or if you want to save 0.5 seconds in the grading (or if you have someone else doing proxys for editing and that kind of stuff), you might as well put "EI 800" before shooting the shot so that the actors don't look too dark x)

 

For Prores/DNX, because there's some compression between the encoding (LogC, this one does take into account what EI you chose on the camera) and the recording of the file, it's a bit more important to choose the good EI before shooting.

[Sebastian Ziegler]

So in that case the native ISO of the sensor is ISO160? (-9/+5) and the image is robust enough to take a 3 stop boost in gain without additional banding and much noise.

 

I guess the difference with other sensors is that at native ISO they don't already have 5 stops over latitude over middle grey.

[Tom Yanowitz]

Yea I guess that's it.

 

Although I'd like it if ISO as a base for exposure was dropped completely for raw digital exposure.

It's just too misleading.

I think ETTR is better personally.

No matter what EI you choose on the Alexa, ETTR will give you the same unique fstop, which is the best one for that particular scene.

("in b4 all the ETTR haters")

If you were to reverse the process and see what ISO would have given that fstop you end up with crazy stuff.

 

For example, if the brightest element of your shot is the face of a white actress, you'll find that the fstop corresponding to ISO 12 will be the optimal one.

That's kind of an extreme almost joke example but you get the idea.

That ISO is not even in the camera and yet it's the ideal one for that particular shot, and might capture shadow detail that ISO 100, 200 let alone 800 wouldn't.

Edited March 24, 2016 by Tom Yanowitz

[David Mullen ASC]

ETTR doesn't make sense for narrative cinematography where consistency of exposure within a sequence trumps exposure of individual shots. This isn't still photography. In a scene, noise levels and exposure values of edited shots have to be within a similar range, within reason.

 

Let's say you have a wide shot with a bright window on one side of the frame, a less bright lamp on the other side of the frame and an actor in the middle. Then in the medium shot the window is not in the frame but the actor and the lamp is. And in the close-up only the actor is in frame. If you apply the ETTR principle to this, the actor would be exposed differently in every set up and in trying to color correct him to a matching level in the edited footage, you'd basically have different amounts of noise on the actor.

 

ETTR only works as a vague principle -- i.e. avoid starving your sensor of light -- but it can't really be applied in practice for dramatic scenes. I mean to really apply it, every time an actor moved around a room in different light levels, you'd be riding the f-stop to keep him at the same exposure level, which makes no sense. If an actor is supposed to step out of the shadows, then he is going to be underexposed in the portion of the shot where he is in the the shadows.

 

Recommending using ISO 160 and ETTR practices might work if you are shooting visual effects plates but it has little to do with everyday working cinematography practices for dramatic fiction.

[Tom Yanowitz]

Well I mean once you know about it you can apply it however you like : scene wise, axis wise, shot wise or intra-shot wise.

 

And when you're giving so much light to the sensor, I don't feel noise variations (say between ISO 50 and ISO 200) on the subject at keylight is that important.

Heck people intercut shots that are not even from the same camera in feature films and viewers don't seem to care as long as what's going on in the scene in good.

 

Of course I'm with you on the awkwardness of what would be "intra shot" ETTR and I never change the fstop during the shot (although some DPs are quiet good at it).

 

But the problem with ETTR is this : the higher end cameras that allow it the most (like the Alexa is perfect fort ir : a little monitor, some LogC monitoring and false color) are also the ones that sort of need it the less because they have a decent range.

If you work with a cheap low-DR cam, ETTR can actually make a big difference and save some shots, but it's harder to apply this method if you can only record and monitor in sRGB or Rec709 (who compress the highlights)

 

The best example I can find for ETTR is this one :

INT NIGHT

The actor is in a room and they are both lit by the same off-screen source.

The source is a lot closer to the actors than the background.

You take a spot reading on the actor : f8 and an average reading on the walls in the background : f1.0.

so if you decide to go with the classical way, you shoot at f5.6 (incending reading near the actor or spot reading of a gray card near him)

So you end up with

- the usual actor at +1

- the background at -5

Unfortunately, at that ISO, your camera is only -3.5/+3.5.

So the resulting image will essentially be a well exposed actor in a sea of darkness.

If that's what you want : great.

If you want some details of the room, you can forget about exposing according to the incident reading, and do some ETTR trickery.

You decide to shoot at f2.8.

- the actor is now at +3

- the background is at -3

So now both the actor and details in the background are being recorded but the actor is way too bright.

Fortunately, you recorded in Raw (or at least a proper log compressed)

So you go in the grading and bring down the highs to where you want them, and you now have both the actor and the background where you want them.

And the only stop that could possibly do that was the ETTR one, f2.8.

 

I wanted to add that in the example that, because it's at night and the DP wants a dark look, he actually goes ahead and underexpose by a stop (f8 instead of f5.6 for the classical exposure), but didn't need it in the end.

 

That last note leads me to the second case for ETTR, my favorite :

Exposing shot with ETTR is the best way to troll on a set.

Your beautiful actress is so close to saturation that when the "pop corn team" of the crew sits in front of a monitor with standard Rec709 monitoring during the shooting, everything looks horrible and clipped.

Then everyone comes to you asking "are you sure about your lighting, don't you want to go check the monitor". That's when you ignore the peasants.

Then at the screening, those same people are sweating in fear of these overexposed shots. So when they see that the graded shots look fine, they understand this one key thing : what the onset monitor shows you isn't anywhere close to what is being recorded.

I don't like people looking at monitors and judging the work while chatting on set and that's a way of telling them.

Edited March 24, 2016 by Tom Yanowitz

[Satsuki Murashige]

Why would you not simply re-light the scene to fit into the dynamic range of the sensor? Diffuse, net, or flag the light where you don't want it. Add more units where you do want it. Result: perfectly exposed image every time.

[Stuart Brereton]

Your beautiful actress is so close to saturation that when the "pop corn team" of the crew sits in front of a monitor with standard Rec709 monitoring during the shooting, everything looks horrible and clipped.

Then everyone comes to you asking "are you sure about your lighting, don't you want to go check the monitor". That's when you ignore the peasants.

.

 

That's when you get fired.

 

Tom, you're a student who has a lot of ideas and research that simply doesn't work in the real, on set world. You might want to call the people watching the monitor 'the popcorn team', but they are the producers, and they cut the checks, and they will not hesitate to replace you if they feel that have to. Arrogantly referring to them as 'Peasants' is not going to help you either, because you are going to encounter these people for your entire career, however long that might be.

 

Here's another thought for you. One of those people watching the monitor might be your beautiful actress's agent, manager or coach. They'll be sure to tell their client just how bad everything looks, and guess what? Yep, you'll be fired. I have seen this happen. And then you get a rep for not being able to light women.

 

We don't do our work in a vacuum. There are other practical considerations in play when deciding on exposure, there are the needs of other departments, and there are political decisions that are, if anything, even more important if you want to remain employed.

[Robin R Probyn]

"I don't like people looking at monitors and judging the work while chatting on set and that's a way of telling them."

 

Tom your really riding for a fall man...if you ever get near a professional set ,even as a loader.. you will be crucified with your attitude.. please listen to professional guys who are being very patient with you .. they are trying to help you..

Edited March 25, 2016 by Robin R Probyn

[Tom Yanowitz]

Guys don't confuse this forum with reality. "pop corn team" and peasants" are just me messing around.

I'm using dark humor and over the top examples to get to the point.

And I've never gotten any complaints about my attitude on set thank you very much.

 

But as for on set monitoring I guess I'm just nostalgic for a time I hardly knew which is the film days.

You made the point "if you expose like this you will be fired"

Let me convert this to a film shoot:

"If a producer looks at your lightmeter and see that the ISO is not the one that he saw a minute ago on the film can, he will not hesitate to fire you"

Cause that's really the same thing.

We criticize the new generation of teenagers for their constant whim for instantaneousness.

But what to make of people who want a good looking image litterally milliseconds after it was captured ?

 

And now, a lot of DPs themselves trust the holy monitor to light and expose.

But unfortunately, some monitors are too forgiving in the shadows.

So he and the colorist cry in silence during post.

And some get fired for that too.

 

I'd rather get fired because the producer doesn't know what the hell he's talking about, than for actually screwing up the image.

 

All that said, you can still work with ETTR and have a fine onset image :

- make some custom LUT/Look

- have a Live-grade connected between the camera and the monitors, like Davinci Resolve Livegrade and grade for the image to look decent on Rec 709.

 

 

 

Why would you not simply re-light the scene to fit into the dynamic range of the sensor? Diffuse, net, or flag the light where you don't want it. Add more units where you do want it. Result: perfectly exposed image every time.

 

Yes of course, but that was an example where you can't.

For example a documentary where you don't have electricians nor equipment

Or a fiction where the actor is on a rooftop and the bakground is now the city, or anything so big you can't light.

Or, another example :

The actor is a child looking at these shadow boxes that make rotating shadow patterns on the walls at night, not sure how it's called.

You can't light the background cause you want the pattern effect both on the child and his room.

Edited March 25, 2016 by Tom Yanowitz

[Stuart Brereton]

But as for on set monitoring I guess I'm just nostalgic for a time I hardly knew which is the film days.

You made the point "if you expose like this you will be fired"

Let me convert this to a film shoot:

"If a producer looks at your lightmeter and see that the ISO is not the one that he saw a minute ago on the film can, he will not hesitate to fire you"

Cause that's really the same thing.

 

 

It's not the same thing. In the film days, no-one ever saw the part of the process between exposure and dailies. If you were overexposing, they never knew, because it had been corrected by the time they watched dailies. Producers didn't know or care much about light meters, they cared about results. These days, they expect to see those results on the monitor in front of them, and unless they trust you implicitly, they're not going to accept sub-standard images, not matter how much you protest that 'things will be ok in post'. You have to understand that most producers DO NOT CARE about technical issues, unless it's costing them money or causing them trouble.

 

On the subject of post, it's very common for low budget movies to have a week or less to color-time. I've had to do some movies in two days. That leaves very little time for finesse, it's broad strokes all the way. In a situation like that, the idea that every single shot will need to be fixed because the DP used ETTR, and consequently the exposures are all over the place, is not going to fly. Assuming you're invited to the timing sessions at all, if you cause them to go over their budgeted time, you will not be popular with producers, and you won't be invited back.

[Ari Michael Leeds]

See Stuart, you talk about being calm under pressure and then you make comments like "That's when you get fired."



He's an inexperienced kid. Doesn't mean you need to flip out on him like a drill instructor. I don't have a lot of respect for that leadership "style" intimidation and beration. And you're not even paying the guy, yet.