Reduced DR at low ISO

[David Mullen ASC]

There's a difference between an underexposed image that is meant to stay underexposed-looking and one that is meant to end up looking normally exposed.

After all, you could have a shot done at 2000 ASA but have part of a room several stops underexposed, rolling into blackness -- that's not really "underexposing" the shot, i.e. rating it even faster -- it just means that parts of the frame are below key exposure.

You could even have the whole shot below key for a dim effect, but it's not necessarily going to get noisier if you never try to lift it up in post but leave it dim. So I don't really consider that sort of thing as defacto rating the camera faster because it's never going to be corrected up to normal brightness (hopefully.)

That test of Leonardo DiCaprio, for example, was done at 2000 ASA but obviously much of the frame is a number of stops below key exposure.

But if you are talking about taking a 2000 ASA shot that was two-stops underexposed and timing it to look normal in brightness, then you have defacto rated it at 8000 ASA.

 

[Roberto Lequeux]

Underexposed meant to stay underexposed at high ISO. I was curious about having large portions of the frame several stops under, but at the same time needing to use a high rating to get the most depth in the frame on night exteriors, where the audience will likely take a close look at those areas.

Since the Leonardo DiCaprio shot is rated at 2000 it is a perfect example. Thanks!

Does anyone remember the lens and stop used?

 

[Jannard]

Underexposed meant to stay underexposed at high ISO. I was curious about having large portions of the frame several stops under, but at the same time needing to use a high rating to get the most depth in the frame on night exteriors, where the audience will likely take a close look at those areas.

Since the Leonardo DiCaprio shot is rated at 2000 it is a perfect example. Thanks!

Does anyone remember the lens and stop used?

Master Prime wide open.

Jim

 

[Eki Halkka]

In Digital Cinema you want to be exposing the sensor about 2 stops lower than you would a "video" camera so the highlights are more filmlike.

Actually, that's what one should do with a "regular video camera" too, if the aim is to gently roll-off highlights (which is usually desirable): expose so that the highs are not blown, use the in-camera gamma etc. settings to bring up the overall brightness to an acceptable level... the main difference is doing most of this in-camera vs. in post (the limits of available DR of course vary a lot between cameras, the "digital cine" ones usually have much more leeway).

 

[Dan Hudgins]

LUT and cameras

LUT and cameras

Actually, that's what one should do with a "regular video camera" too, if the aim is to gentry roll-off highlights (which is usually desirable): expose so that the highs are not blown, use the in-camera gamma etc. settings to bring up the overall brightness to an acceptable level... the main difference is doing most of this in-camera vs. in post (the limits of available DR of course vary a lot between cameras, the "digital cine" ones usually have much more leeway).

Well, today video cameras have a LUT used before compression, but "classic" video look was from using cameras without built in LUT going to limited range FM recording on video tape, it was mostly the broadcast limits or the FM tape limits that give it the "video" look, and not being gamma 0.4545 in the camera to compensate for the gamma 2.2 monitors.

Todays high end video cameras are like Digital Cinema Cameras, in that both use LUTs between the sensor data and the final graded RGB image. Cameras like HDSLR and other camcorders use their LUT before recording because their compression is of limited bit depth, REDCODE being 12bits it seems lets the LUT be used after redording. No camera with LUTs and a modern low noise sensor that works right has the classic "video" look, the "video" look will be a thing of the past as soon all camcorders have built in LUTs and full gamma correction. Some of todays broadcast cameras seem to un-correct the gamma to reduce shadow noise, similar to "crushing the blacks" to un-do some of the shadow gamma correction.

How much any camera can "underexpose" the sensor depends on the sensor size, the moving pattern noise which can be above the bottom bits, the fixed pattern noise and so on. You can pipe a sensor through a 14bit ADC, but you may see vertical lines in dark shots or in the sky etc. because the fixed pattern noise it above the bottom bit. As sensors get better and the sensor boards let less hash into the preamps, readout timing, and ADC circuits, you will see more cameras with higher ISOs and better highlights since most of the image will be in the bottom 6 bits (LSB) rather than the top 6 bits (MSB).

Still, if you look at the images 1:1 pixel, you should see some banding if you have the ISO high enough to shoot with 18% gray card at the 4th bit from the bottom, and have the gamma correction putting that 18% gray card at 0.458 in the finished gamma 2.2 corrected RGB image to look at on a monitor.

If you reduce a 4.5K RAW image to a 1K image you get more tones, and so its like using more bits in the ADC, so if you are making a DVD you can shoot maybe two stops higher EI/ISO speed than you could making a Blu-Ray because there are more pixels to average.

I noticed that in some information about the curves used in ALEXA the heavy S-curve they seem to be using at higher ISO do not give even spacing of tones to lower values as near mid-tone, that can hide noise and banding, but does not make the shadow contrast like a well exposed ECN negative for subject matter that is the same light reading down from 18% gray card held in front of the primary subject as in normal photography, maybe. If you make the S-curve very pronounced to hide the fact that you have 18% gray card subject brightness 6 bits from the bottom in the exposure, there have to be some histogram gaps (in an 8 bit RGB output) unless you have "de-noised" the data in some way to interpolate where the tone steps should be (since the S-curve pushes some of the lower bits down, it makes the space between the bits close to the 6th bit wider, if the S-curve is heavy enough you get gaps going to 8 bits, if less then they show up going to 10 bits for display). Its more like looking at a pushed film negative where the shadow tones are not on the straight line part of the films transfer curve. You can fix that in post, but you will then see the banding and noise and more histogram gaps, which is why its better to add fill light at the time you shoot. Since the bottom two bits have heavy noise (random, fixed pattern, and moving pattern), the top ISO for any sensor+ADC is with 18% gray card exposure no lower than the 6th bit or so. Otherwise the EI/ISO numbers have no meaning if the 18% gray card's exposure on the sensor is not lifted to midtone of 0.458 (116 on an 8 bit display). If you display the 18% gray card at values lower than 116 on an 8 bit display, then you are exposing it at lower ISO numbers as any other meaning of EI/ISO is absurd. If you put the 18% gray card in a shadow in the frame so that it ends up under 116, then it is underexposed at the processed ISO, if you put it in a bright area of the frame and it end up over 116 then it is overexposed. That has nothing to do with what looks good as far as cinematography goes, it is the choice of how the image ends up, but to give EI/ISO numbers meaning, at least midtone should end up right which is why Kodak had the LAD test film, to put midtone in the right density on the prints, if you want the subject under mid tone, it does not mean that they are "exposed" as some high EI/ISO number it means they are underexposed for the stated EI/ISO number.

 

[David Rasberry]

In a 12bit sensor/data you do not have much latitude, since if green clip is 160 (90% white card close to ADC max), then one stop down is 320, two stops down 640, three stops down is 1280, four stops down is 2560 (at four stops you have only 8 bits of "data+noise" so you see banding). You need 10 bits to avoid histogram gaps and banding after grading, so that gives you a speed of 320 for a 12 bit sensor. (the ISO curve expands the lower tones so 10bit linear data is close to 8 bit data at gamma 2.2 for banding in the shadow areas).

Thanks for these posts Dan. The above excerpt neatly explains the rationale for a "native" 320 ISO rating for a 12 bit sensor.

When a few of us were getting a hands on demo with an R1 from Peter Majtan, one of the things we observed was that Peter always exposed for 320 and used a 320 LUT as his monitoring guide. Didn't matter whether he was shooting an indoor tungsten scene, a high contrast daylight outdoor scene, or a low light sunset or night time street scene. He always got consistent and excellent results with this approach, the raw files gave a lot of range for post manipulation.