ETTR: The Ultimate Exposure Technique?

[Ebrahim Saadawi]

Some old tests I dug out on Exposure. I conducted the same test on many platforms, Canon 60D/5D mk III/Canon C100 Clog/Nikon D5300/D810/Sony A7s at SLOG2, and Panasonic GH4 wat CineD & CineV and even did a quick on-set test on an Alexa, shooting 2.8K arriraw, and ALL these camera systems give the exact same results as follows:

ETTR Log ungraded

"Correct exposure or slightly under" Log ungraded

ETTR corrected to match the correctly exposed

ETTR

Correctly exposed

ETTR to correct exposure

I mean it's such a huge difference in image quality that I never expose correctly or slightly under anymore, the higher you go in exposure and the more you feed the sensor light the better quality, thicker colour, less noise and more DR you get, to my eyes anyway.

It even works if you achieve ETTR by pumping up the ISO/gain (but it's best achieved by opening iris/lowering shutter/increasing light), a 320ISO underexposed image shows way more noise than a 3200ISO overexposed image, so it's underexposure that creates noise, not high ISOs!

Exampe for 320ISO (under) vs 3200ISO (over):

320ISO

3200ISO

Here's both when you grade them to normal point:

The first noisy one of the two graded images above is at 320 ISO, while the bottom clean one is at 3200 ISO, with the same settings/light and everything, yet the 320ISO shows significant noise because the image is underexposed, while using the ETTR technique the I can get perfectly clean images at 3200ISO using a Canon rebel (as seen on last image, at 3200ISO with a Rebel)

again to my eyes, It's underexposure that creates noise, not high ISOs.


Plus, ETTR increases Dynamic Range (as long as you don't clip the HL) because once you start dialing down exposure, the highlights don't gain anything while the shadow noise floor is increased = lower DR, so with ETTR you're using the entire DR of your sensor, squeezing every last drop of performance)

Achieving ETTR is best done using a Waveform monitor or an RGB Histogram in-camera, you increase exposure until the image touches the upper boundary (WF) or right boundary (Histo), and Zebras if set to 100% could be used too, you increase exposure until they show up, then back down one notch. Yes the image looks ugly, flat and over exposed at first, but that's the way we shoot now anyway in a Log/LUTs dominated world. So this technique is only viable if you are doing correction to the image in post and not handing them off for direct delivery obviously, for these type of jobs get a "correct" exposure with skin at 70IRE and that stuff...

So what do you guys think? I see many professionals don't expose that way so there must be a counter argument, lets hear it please!

Or can I just use this simple technique for setting exposure and rest my mind in peace after all these years struggling to get "correct" exposure?

 

[FrameFarmMedia]

With over exposing you risk losing detail in the most compressed part of the gamma curve would you not?

I don't understand how you came to the conclusion that you increase DR.

 

[DPStewart]

^-- yes, you're correct.
If you look at the swaths of fabric in the background-right of the shots above you see really clearly the lower DR and the more harsh clipping - especially on the white fabric.

But exposure is ALWAYS a trade-off. You have to employ this technique on a scene-by-scene basis. There will be many shots where the cleaner bulk of the image may work better despite the lower DR. And vise-versa.

In the above pictures (the last 2) I would prefer the look of the ETTR final image in almost all cases.

 

[Ebrahim Saadawi]

About the white fabric, it was actually clipped on the waveform so my fault not ETTR. What I mean exposi highly WITHOUT clipping the highlights at all, this gives me the highest quality.

The counter argument for ETTR is that the highlights are the most compressed areas of the image, thus theoritically you're pushing the imahe into the most compressed areas, getting more artefacts and break up of the image, but the thing us, it doesnt. Not in my testing, from what I can see it's actually the lower end of the codev that contains the most codec artefacts and pushing the image closer to that area significantly decreases image quality.

Where higher DR comes from: imagine this, you're shooting a interviewee in a room, and there's a bright window behind him in the frame

-If you undersexpose below the window clipping point, where detail shows outside, the
lower you go in exposure, the more noise you get inside the room (the noise floor increases)

-If you ETTR, to the point just below the window clipping, this way you're gettung the most information in the room that you can get while still holding the window just as example 1

so more shadow information with cleaner shadows, less artefacts, while not losing any highlight detail too.

By increasing exposure and pushing down the ISO, you're directly affecting the signal/noise ratio, and getting hugher DR.

exposing lower than where the highlights clip, the lower you go, the more noise you get in the shadows, the lower the DR.

That's how the theory works, but moat of all it works in real life tests!

 

[Grug]

I'll take good tonality and highlight headroom over noiselessness any and every day of the week.

The only format that I believe it makes sense to expose to the right on is linear raw recording. For log recording it completely defeats the purpose of recording in a logarithmic gamma, and you end up with godawful plasticy skin tones.

 

[Ebrahim Saadawi]

Yes it's the only counter argument I'm seeing on the forums, that you somehow lose skin tone detail and highlight 'tonality' when exposing them up the curve, which is the same argument that highlights are more compressed parts of the codec,
Fortunately I have some free time so I am testing for skin tones and highlight tonality today, to see if I lose any when ETTR and adjusting in post using an 8bit codec, will post back
Other tests are HUGELY welcome. There isn't much information about this on the web other than theory
We've already established that ETTR gives much cleaner images on terms of noise, now lets look at ''highlight/skin issues''. You can simply shoot someone with a high expsoure (putting his skin into the supossedly most compressed part of the codec) then shoot the same with a darker normal exposure and compare if the adjusted ETTR skin shows less information or more break up. Easy test really.

 

[Joshua_G]

Thank you, Ebrahim.
It should be extremely meaningful and useful test, to see whether midtones and highlights tonality is being suffered by ETTR, after grading.

 

[abreu-canedo]

I'm shooting RAW with the FS700 into the Odyssey 7Q...which is a bit tricky because it's RAW, yet, the camera is sending Slog2 signal which I believe is burnt into the RAW signal. (correct me if I'm wrong)

With that said, I did a shoot where I significantly overexposed one shot (skin at 50IRE), then shot it again with skintones between 32-38IRE on my false color.

Maybe there are more variables to consider here, but I actually had a really hard time grading the skintones on what I thought was the "properly" exposed image, and ended up thinking that the overexposed image actually gave me better skintone. Here is the final result (poor quality screenshot, but still can see that skintone came out well):



I'm still new to exposing RAW on my specific FS700/7Q set-up, though...so there is still a lot to learn.

 

[Joshua_G]

Hi Rafael,

When shooting in RAW, whatever parameters are set in the camera, including log, are included as an informative data added to the image file, separate from the image data, however the data being outputted by the sensor isn't altered or manipulated in any way. The informative data which accompanies the image data only tells the software dealing with the image data (conversion, editing or grading) how to interpret it.

As for the degree of ease, or difficulty, in grading colors, including skin tones – this by itself is totally unrelated to the question about the degree that tonality details are being kept at various levels of exposure (i.e. 'correct' exposure vs. ETTR).

 

[abreu-canedo]

Hi Joshua,

Thanks for the explanation on the Slog as metadata. This explains the results I'm getting. I'm going to be using ETTR from now on when shooting RAW, or at the very least, over-exposing by a stop or two. Would this be sound practice?

When speaking of degree of ease towards grading pleasing skintones, I mean that I found my ETTR shots to end up giving me better skintones when graded (not plasticky). And when I was exposing my RAW footage the way that I thought I should be exposing Slog2, that I was not getting the skintones to look as nice in the grade (I spent more time, whence more dificult). One isn't harder than the other, but I found them to give different results and the ETTR shot to be more pleasing. Difficulty just relates to the amount of time I spent trying to get the best look.

Thanks again for the good info!

 

[Joshua_G]

Hi Rafael,
There may be a confusion between 'correct exposure', 'over exposure' and ETTR.
It's after midnight here, so I'll elaborate about it tomorrow.

 

[abreu-canedo]

Sounds good.

I'm familiar with ETTR (without clipping). And when I say "overexpose", I mean one or two stops above what a light-meter reading would indicate (again, without clipping). But yes, please feel free to clear up any confusion tomorrow.
-Thanks again, Rafael

 

[Grug]

Sounds good.

I'm familiar with ETTR (without clipping). And when I say "overexpose", I mean one or two stops above what a light-meter reading would indicate (again, without clipping). But yes, please feel free to clear up any confusion tomorrow.
-Thanks again, Rafael

Half a stop over is one thing, a full stop overexposed is another. But two full stops overexposed? That just seems to be asking for trouble.

 

[Joshua_G]

Sounds good.

I'm familiar with ETTR (without clipping). And when I say "overexpose", I mean one or two stops above what a light-meter reading would indicate (again, without clipping). But yes, please feel free to clear up any confusion tomorrow.
-Thanks again, Rafael

The in-camera light-meter reading is far from being accurate, it considers the frame to be 'an average-lit, average subject' – which is rarely the case.

 

[MichaelFischer.at]

Half a stop over is one thing, a full stop overexposed is another. But two full stops overexposed? That just seems to be asking for trouble.

Hi Grug,
that really depends on the gamma curve that you're shooting on. If you take S-Log2, the most information is held in the first stops below clipping:
http://community.sony.com/t5/image/serverpage/image-id/7925i082E029E2F42EFAD?v=mpbl-1

Currently I'm shooting an A7s with S-Log2. I rate it at ISO800, that means overexposing by 2 stops. S-Log2 has a huge highlight range (6 stops above mid grey), therefore adding two stops brings it back to the usual DSLR headroom of 4 stops above mid grey.

My post workflow is like this: S-Log2 => Linear => -2 stops => S-Log2 => r709 1D LUT. This is almost like shooting raw and I get the exact same results compared to exposing normally (without the noise ).

 

[Grug]

Hi Grug,
that really depends on the gamma curve that you're shooting on. If you take S-Log2, the most information is held in the first stops below clipping:
http://community.sony.com/t5/image/serverpage/image-id/7925i082E029E2F42EFAD?v=mpbl-1

Currently I'm shooting an A7s with S-Log2. I rate it at ISO800, that means overexposing by 2 stops. S-Log2 has a huge highlight range (6 stops above mid grey), therefore adding two stops brings it back to the usual DSLR headroom of 4 stops above mid grey.

My post workflow is like this: S-Log2 => Linear => -2 stops => S-Log2 => r709 1D LUT. This is almost like shooting raw and I get the exact same results compared to exposing normally (without the noise ).

Hi Michael,

I'd be very interested to see some of the results you're getting from that methodology. It flies in the face of 'correct' exposure for a logarithmic gamma by placing important things like skintones up in the more compressed part of the curve (which becomes even more limited when you're only recording in 8-bit).

I'll be genuinely surprised if that methodology isn't getting you same plasticy skintones I'm seeing on the majority of A7S SLOG2 footage at moment.

Cheers,

Mark

 

[MichaelFischer.at]

Hi Mark,

just FYI, I was contributing the log (S-Log2, C-Log, Log-C) formulas for MlRawViewer. Below you can find a plot of what these curves are doing vs. the linear data. The chart is mapped to middle grey on Canon 14 bit linear DSLRs. These cameras have mid grey at 1024 (-4 EV). If you can wrap your head around what the charts mean, you will know why the A7s has to have a native ISO of 3200 in S-Log2 and why it is so noisy. Just take a pic with your SLR at -2 and then apply +2 in Lightroom, same thing.

Fact is, the more you expose, the more linear data you will get. Also the logarithmic formats have more code values the higher up you are, or, the closer you get to clipping. So all this contradicts the statement, that you should expose correctly. Even Alister acknowledges this recently:
http://www.xdcam-user.com/2014/08/e...-on-the-sony-a7s-part-one-gamma-and-exposure/

If you only have a fixed LUT, then you really have to stick to the given rules. Since I'm converting log to linear (which is very easy mathematically) and adjusting exposure that way, it's almost impossible to get different results. Again, I already did that, just have to fetch the screenshots if you are interested.

Now the fun begins if you map log to r709. If you do that only via lift/gamma/gain or S-curves, you are compressing certain parts of the image. Skin can look very flat. If you use a proper log to r709 LUT you should see a big difference. You might have to protect your highlights with a knee afterwards, as the gamma function is very prone to clip, but you will get a nice gradation in the skin tones.

Hope that makes sense

BR
Michael

S-Log2 (purple) vs LogC

 

[Razz16mm]

Example ETTR from my D16. Linear uncompressed 12 bit DNG. Shooting log curves essentially is compressing exposures to the right already, so the technique is more limited. I have recorded ETTR exposures unclipped that push virually all values off REC709 referenced scopes in Resolve and recovered them, though that extreme does compromise tonality. The D16 has a direct raw view mode, similar to Red, that let's you see what is clipping and what is not down to the pixel. Great for critical exposure judgements. This is a 1 stop over exposure.

 

[egproductions]

The more you expose to the right the less you can bring back from the right. Displays are 8bit so If you take an 8bit image and expose to the right you are leaving yourself with less than 8 bits of data. In some situations that might be ok, in many it will not be.

If you are slightly over exposed you can save yourself and try to pull highlights back, if you ETTR then you will not have that saftey.

When shooting 12bit raw files, you can definitly get away with ETTR in most cases and yes it will yield cleaner images.

 

[Samuel H]

I tend to ETTR if there's nothing in the image that's at risk of clipping (which means: not that often, actually), and also trying to keep a constant amount of overexposure in all shots (otherwise you end up with a lot of work in post and varying amounts of noise from shot to shot).

I then use my a7S+slog2 Premiere exposure adjustment curves. Quick example of how well they work: