(Auto) ETTR (Exposure to the Right): -- History & Beginners Guide

[RenatoPhoto]

Looking for a little information about Exposure to the Right, here are some articles that will help to explain the benefits of this exposure technique.

http://www.luminous-landscape.com/tutorials/expose-right.shtml
http://www.luminous-landscape.com/tutorials/optimizing_exposure.shtml

The luminous-landscape information was provided thanks to Michael Reichmann.


Loading the module

To use this feature, you must first load the module in the module tab:



After restarting the camera, you will now find the module located in the Expo tab:




Using the module

First, lets take a look at the default options:




Trigger Mode:

• Always On:  AETTR is always active.  In photo mode through the viewfinder, this means that AETTR is always inspecting the exposure in the image review, and setting the exposure settings for the next shot.  In Live View, this means that AETTR is always looking at the image in the live view, and adjusting the exposure settings as necessary.
• Auto Snap:  After taking a shot, AETTR will take another shot if necessary.
• Press Set:  AETTR activates by pressing the SET button.  In photo mode through the viewfinder, AETTR will turn on live view, determine the correct exposure settings, and return to photo mode.  In live view modes, AETTR will determine the correct exposure settings.
• HalfS DblClick:  This functionality is exactly like Press Set above, except the trigger button is half pressing the shutter button twice.

Slowest shutter: 

• Set the slowest shutter speed that AETTR is allowed to use.

Highlight ignore:

• Set the amount of pixels in the image, that are allowed to be overexposed.
• This works like an specular highlight control.

Midtone SNR limit:

• Set the minimum Signal to Noise Ratio of the median brightness level.
• "The median "brightness" gives you the signal level, so we can use the SNR. So, when I say the midtone SNR is 5 EV, this means half of the image has a SNR less than 5 EV, and the other half has a SNR higher than 5 EV. Pretty easy and statistically robust".
• http://www.magiclantern.fm/forum/index.php?topic=5200.msg70012#msg70012
• Note:  This setting by itself, overrides the overexposure control of AETTR.  This means, that AETTR will overexpose the image as needed, to hit the SNR limit set here.

Shadow SNR limit:

• Set the minimum Signal to Noise Ratio of the shadows.
• "For shadows, I've chosen the 5% percentile. So, in this context, the shadows having a SNR of 3 EV means 5% of the image pixels have a SNR lower than 3 EV, and the other pixels will be brighter than that".
• http://www.magiclantern.fm/forum/index.php?topic=5200.msg70012#msg70012
• Note:  This setting by itself, overrides the overexposure control of AETTR.  This means, that AETTR will overexpose the image as needed, to hit the SNR limit set here.

Link to Dual ISO

• Let ETTR change Dual ISO settings so you get the SNR values in mids & shadows.  It will disable dual ISO if not needed.
• With the SNR limits above, instead of simply allowing overexposure to reach the set SNR limits, this will enable dual ISO instead, allowing you to retain full highlight detail (minus dual ISO deficiencies), while still capturing the required SNR in the midtones and/or shadows.

Advanced settings




Exposure target:

• Where to place the highlights with respect to overexposure.

Allow Clipping:

• Off:  Only allow overexposure as per the highlight ignore setting.
• Green Channel:  Ignore the green channel, and calculate exposure settings from the other channels.
• Any Channel:  Allow some overexposure, over and above the highlight ignore setting, in any channel.

Link to Canon shutter

• Set the slowest shutter speed AETTR is allowed to use, via Canon exposure settings.
• This allows you to set the slowest shutter speed with the scroll wheel/main dial.

Show metered areas

• Show where the white point and SNR levels are metered (what exactly is considered highlight, midtone and shadow).
• Colored zebras represent those areas.

Show debug info

• For nerds.

[Audionut]

Hey mate, good job.  I was going to have a go at this, but you did a much better job.  I'll add some more information as time permits.

A couple of things to note.

ETTR BENEFITS SUMMARY: The great advantages of ETTR, in case that all of this reading was way above your head here are the potential benefits:

1. Reduce the noise in the underexposed areas.
2. Capture more detail and less noise.  i.e. The amount of usable information (signal) is higher than the noise, improving  the Signal-to-Noise Ratio.
3. In most cases ETTR allows to use lower ISOs since more light is reaching the sensor for example by lowering the shutter speed.
4. ETTR in RAW mode is more accurate and the implementation of histograms in jpeg.
5. Also posted by one of our colleagues, there should more color accuracy as noted here:
http://www.magiclantern.fm/forum/index.php?topic=5149.msg31285#msg31285

ETTR does not capture less noise.  The logarithmic curve of SNR remains the same for all ISO's.  We are merely pushing the signal to the right of the curve where the SNR ratio is greater.  The noise however, has remained the same.

In most cases, ETTR requires a higher ISO.  ETTR requires a longer exposure time, if you are already at the minimum shutter and aperture that the shot requires, a higher ISO is needed.
At POTN, they call it HAMSTTR to take into account the ISO changes.

[noisyboy]

Great post man!

[RenatoPhoto]

@ Audionut

I followed the epic HAMSTTR thread and wanted to proof to myself that lower ISO is better for ETTR so I setup a test where the camera calculated the ETTR amount using the ETTR hint provided in the histogram.  According to the algorithms programmed by A1lex the ETTR hint tells you how many EV you need to change the exposure to reach optimum ETTR. 

So first I setup the ETTR at the following values i.e.

ETTR percentile 98%
ETTR target level -05. EV
Clipping mode No clipping

Camera 5D3 with 180mm f/3.5

First image (top right) is in AV mode to hold the aperture at 6,3  and ISO in AUTO MODE.  The camera chose 1/200 with ISO 1250.  At this settings the ETTR hint from ML calculated that I could have overexposed by +1.9 EV

In the next images I set the camera in manual mode and found some settings at different ISO where the ETTR hint was close to zero.  This would mean that for those settings the image was correctly exposed based on ETTR theory.  I went from ISO 4000 all the way to ISO 100.

The images were processed with Photoshop and all were set to  Auto exposure so tha ACR would find its optimal level of correction and have all images looking the same.  Then I cropped at 400% to show the noise and small details.



It is quite clear that lowering the ISO also greatly lowers the noise.  It is also quite amazing how much noise can be removed by the ETTR technique by comparing the first image with the one below at ISO 1600.

Next I enabled the ETTR feature and set my Slowest shutter at 1/2.  Took some shots an noticed that ETTR ON it did not immediately drive the settings towards ISO 100.  With the current algorithms the ETTR system will change the exposure to obtain E=0 but it will not change ISO set manually in the camera unless the required EV change should be more that +- 3 EV.

For example if I manually changed the ISO to 4000 and the shutter to 250, which the camera interpreted a properly exposed by looking at the Exposure Level indicator = 0.  The ETTR system would only change the shutter speed to 160 and leave the ISO at 4000.   Now if I set the speed so that my Exposure Level indicator >= 3, the the ISO would change.  For example set the speed to 1/40 (EL=3) and the I
set the ISO=4000, the camera would automatically lower the ISO to 1000 and leave the speed to 1/40 which showed E=0.

Conclusion 1 = Lower ISO produce better ETTR images that higher ISOs.

Conclusion 2: The ETTR system is doing a great job but it is not exploring the possibility of lowering the speed all the way down to my setting of 1/2 sec.  Some people, such as those found in the HAMSTTR thread do believe that higher ISO is better, but IMHO they are partially wrong.  If you cannot lower the shutter speed then you must raise the ISO to obtain proper ETTR, and in this case a higher ISO is better.  But if you can lower the shutter speed and reach a lower ISO, and do the same proper ETTR at that lower ISO, the image will be much better.

[Audionut]

Conclusion 2:  If you cannot lower the shutter speed then you must raise the ISO to obtain proper ETTR, and in this case a higher ISO is better.  But if you can lower the shutter speed and reach a lower ISO, and do the same proper ETTR at that lower ISO, the image will be much better.

This is correct, but I think your understanding of what I was explaining is flawed 

When ascertaining the results of shutter/aperture/ISO (exposure triangle), only 1 option can be examined at a time, the other 2 must remain the same.

ETTR works by increasing exposure to the sensor saturation point.
So by saying the lower ISO is better, and showing the results of this with a slower shutter speed also, the result is flawed, as the lower ISO also has the advantage of a longer exposure from a slower shutter speed.

A true test of ISO (in ETTR) is to have a set shutter and aperture that doesn't change.
Below is a quick test showing my dirty table.
All shots were with a shutter of 1/15 and an aperture of f/4.0.  Note:  A minimum shutter and aperture must be taken into account.  In the field, we cannot just set whatever shutter/aperture we want, as we are limited by our DOF requirements and minimum shutter if handheld etc.

note:  The post exposure results are direct from ML and performed in post, and the image sizes are the crops.

ISO 100 - post exposure +2.25EV - 211.05KB


ISO 200 - post exposure +1.26EV - 174.39KB


ISO 400 - post exposure +0.22EV - 154.07KB


ISO 800 - post exposure -0.77EV - 146.94KB


ISO 1600 - post exposure -1.79EV - 143.27KB


ISO 3200 - post exposure - -2.79EV - 142.42KB


And some examples from a copped area that was further in the shadows.
Exposure results are the same.

ISO 100 - 212.03KB


ISO 200 - 171.75KB


ISO 400 - 148.30KB


ISO 800 - 143.48KB


ISO 1600 - 135.01KB


ISO 3200 - 134.02KB



Some more points to note.
Some ISO's are underexposed in camera and increased in post as I wanted a minimum shutter of 1/15.  1/15 is below the handhold capability of most body/lens/user combos, but felt is was a good speed (with the available light I had for the test) to highlight my point.
Of course a slower shutter would have been better, but it highlights that sometimes you must use a higher ISO for correct exposure, you can't just keep slowing the shutter.

Conclusion.

The shadows are where all the action is!  The highlights have enough Signal To Noise ratio at almost all ISO's.
It gets hard to see the differences in noise in some of the test results.  Hint:  Look at the file sizes.  They keep reducing the higher the ISO (less noise).

Note, the results below reference a Canon 5D3.  The results hold true for all cameras, it's just that other bodies will have different SNR results for different light levels.
The SNR of an ISO bump reduces by 3dB for almost all ISO's.  Reference this DXO chart.


In other words, the noise doubles for every doubling of ISO.

Now reference the SNR of all ISO's at all light levels.


Here we can see again, that in the highlights, even ISO 6400 still has a SNR of 30dB, which is greater (by 6dB) then the SNR of ISO 100 @ 1% light level.

And here is where it gets interesting.  If we have an exposure @ ISO 100 that results in our shadows being around the 0.1% light level, we can see that we only have 6dB of SNR (ouch).
Now if we take the same scene and increase ISO to 200, we double the exposure, so that 0.1% light level now lies at 0.2% exposure on the Full SNR scale.
ISO 400 0.4%
ISO 800 0.8%
ISO 1600 1.6%
ISO 3200 3.2%
ISO 6400 6.4%
And at 6.4% light level, ISO 6400 has a SNR of around 18dB vs the 6dB of the same shot at ISO 100.  Yes, the highlights and midtones will have a higher noise level (lower SNR), but our shadows (where lets face it, all the noise actually is, or more correctly, where the noise is noticeable due to low SNR) OMG.  We gained 12dB of SNR just from using a higher ISO.  And these results are reflected in the sample images above.

Yes we can lower shutter speed and/or reduce the aperture to achieve the same results, and if the situation allows you to do so, this is the best way to ETTR, but when you can no longer reduce shutter and/or aperture, hopefully I've shown that increasing the ISO can actually be beneficial.  It's not simply a case of higher ISO = more noise.
It's better to increase ISO to gain the best exposure, then to increase exposure in post.

And herein lies the advantage of ETTR.  Bring those shadows out of the noise (higher SNR) and then as an added benefit, reduce the entire exposure (and the noise ) in post.  Win all the way.

[garry23]

There is a lot of physics going on here. I commend these two articles to those who wish to have a greater understanding of their sensors:

http://www.clarkvision.com/articles/iso/

And for the 5DIII owners

http://www.clarkvision.com/articles/evaluation-canon-5diii/

Cheers

[ilguercio]

Very good thread.
I have been a huge supporter of ETTR during the years as it's another way of making the most out of our cameras.
@audionut: sorry for breaking the thread flow but apparently (from the cigarette package i see) you are from Australia.
What area?

[Audionut]

I commend these two articles to those who wish to have a greater understanding of their sensors:

Excellent read.  Thanks.

What area?

SEQ.  Not Brisbane

[newsense]

I'm sorry but after reading this thread and all the linked articles I think I am even more confused. If I understand correctly ETTR only makes sense if shooting raw and not 8bit, is that correct? Since I am not shooting raw where does that leave me, ETTR , expose in the center or ETTL and does picture style make any difference when setting exposure? If I am using a "neutral" or "flat" picture style would I set my exposure using that picture style or set exposure with a standard style and than shoot "neutral"? Thanks in advance.

[Audionut]

@newsense.

When shooting in JPG or H.264 where post processing is destructive, I would recommend shooting with the correct exposure (0EV on the meter) in camera.

[3pointedit]

Excellent thread! I guess this is why cameras like the Red uprate their preffered operating ISO? Around 800ISO at the moment from memory.

[bart]

@audionut

Very nice info. Why not make a frontpage article from this nice info and show this great new feature to the public. I can help you with that if needed.

[Audionut]

Thanks bart.  It was RenatoPhoto who got the thread started with another excellent OP.

My biggest issue with tackling something like this is making it legible for new users.  And of course, finding the time to compile all the information.  It's a rather large topic that requires some brief descriptions of lots of related subjects.  I'll ask DXO if we can use a couple of their graphs, if they answer yes, it will give me some incentive to get in and get it done 

[bart]

Yes I see that now.  And I see RenatoPhoto, squig and you are doing a fine job with forum maintenance and build testing. Very nice.

[Freelancer]

Thank you for the information. Great job!
I tried to activate this feature on my 5D3 but it is not available in the Expo menu. Any idea?
Does the feature work in the RAW video mode as well?
Thanks for the help.

[Marsu42]

And at 6.4% light level, ISO 6400 has a SNR of around 18dB vs the 6dB of the same shot at ISO 100.  Yes, the highlights and midtones will have a higher noise level (lower SNR), but our shadows (where lets face it, all the noise actually is, or more correctly, where the noise is noticeable due to low SNR) OMG.

Interesting read, thanks, but for your general conclusion, I'm not so sure. I understand you're saying that a higher SNR in the shadows is more important than in the midtones (where do the midtones start?). I say this heavily depends on the scene, because probably the shadows won't be important at all, in the background or even clipped by postprocessing to get more punch, while noisy midtones might bother much more if they are smudged by noise reduction.

Also the color rendition goes downhill with higher iso, that's why I'm still not so sure if a properly/ettr exposed shot @iso1600 something on the 18mp sensor is always better than iso800 underexposed. Plus of course if the lighting changes you'll quickly run into blown highlights when you're exposing for ettr, while some underexposure gives a safety margin.

[newsense]

@newsense.

When shooting in JPG or H.264 where post processing is destructive, I would recommend shooting with the correct exposure (0EV on the meter) in camera.

Not arguing your point but what uses do the waveform and histogram have if canon's EV meter is the best exposure tool? I just shot 4 videos. One exposing at 0EV on Canon's EV meter, one using ETTR making sure not to clip any RGB channels or luma using ML histogram, one using ETTR making sure not to clip luma using ML histogram and one using the ML histogram and exposing in the center. All were shot using neutral PS and before CC they don't look that much different besides the one where I used ETTR and made sure not clip any RGB channels, that one looks pretty underexposed. Maybe after CC I can come up with the best way to expose.

[Bioskop.Inc]

WOW! Thanks for all this information its been so helpful for video (i'll get to using it for pictures soon).

It really improves the footage!

[Audionut]

Not arguing your point but what uses do the waveform and histogram have if canon's EV meter is the best exposure tool?

It's not.  But if you want to limit yourself to JPG/H.264, them's the brakes.

where I used ETTR and made sure not clip any RGB channels, that one looks pretty underexposed

You should brush up a little on exposure and how to read a histogram.  There was a strong highlight in your scene that was triggering the overexposure warning in the histogram, causing the overall scene brightness to be low.

If you can retest with the same scene, enable zebras so you can see exactly what is being overexposed.  Use the same settings as the Canon exposure meter (0EV) and have a look at what's being overexposed.  Determine if that highlight is important.  If it is, then technically, ETTR was not underexposing, the Canon metering was overexposing.  If that highlight is not important, keep adjusting exposure until an important highlight does start overexposing.

[newsense]

It's not.  But if you want to limit yourself to JPG/H.264, them's the brakes.

You should brush up a little on exposure and how to read a histogram.  There was a strong highlight in your scene that was triggering the overexposure warning in the histogram, causing the overall scene brightness to be low.

If you can retest with the same scene, enable zebras so you can see exactly what is being overexposed.  Use the same settings as the Canon exposure meter (0EV) and have a look at what's being overexposed.  Determine if that highlight is important.  If it is, then technically, ETTR was not underexposing, the Canon metering was overexposing.  If that highlight is not important, keep adjusting exposure until an important highlight does start overexposing.

Thanks for responding. Exposure seems to be a problem for me since I got my DSLR. Mainly due to a lot of conflicting information i've came across. I would shoot RAW but I only have a 650D so I am limited to H264 for now. I guess my biggest misunderstanding is should I expose based on luma or RGB? I read somewhere that even if luma channel isn't clipping individual RGB channels clipping can throw away color data. I understand what you mean by only worrying about important highlights clipping, I generally ignore clipped bright windows. From your response I assume you suggest ETTR based on readings from the ML histogram rather than the canon EV meter? Sorry if my questions seem a little noobish but I am fairly new to DSLR and thanks for your time.

[Audionut]

@Marsu42

First of all, I'd like to point out that life is a compromise.

I honestly am not sure, on how best to communicate with you.  So I'll just be me, and if you take offense at my responses them's the brakes.

I say this heavily depends on the scene, because probably the shadows won't be important at all, in the background or even clipped by postprocessing to get more punch, while noisy midtones might bother much more if they are smudged by noise reduction.

Generally (most of the time) my exposures are in the higher part of the histogram (ETTR for the win), as I generally don't shot scenes with a large dynamic range, in which case, yeah, I don't give a shit about the shadows either.

But I respect that some other people do appreciate having clean shadow detail.

I explained reasonably clearly I thought, how the Signal to Noise Ratio worked through the exposure.  It makes sense (to me at least) that if you value your midtones more importantly, you would adjust your settings appropriately.  The same goes for,

Plus of course if the lighting changes you'll quickly run into blown highlights when you're exposing for ettr, while some underexposure gives a safety margin.

If you place a higher priority on your highlight detail, you would adjust your settings appropriately in this situation also.

Also, I used the shadows in my example as they show a greater benefit with ETTR.  Midtones also show a benefit with ETTR.  This much should be obvious in the samples crops in my post above.  Those crops aren't of extreme shadows  

You're trying to find flaws in my specific examples referencing all manner of different situations.  A little of the grey matter is needed.

This isn't a guide on exposure, histograms, how to take a photo.  It's a guide on how to push your exposure to the right, and the benefits that can be obtained through this practice.  Some prior knowledge of exposure is needed.

Also the color rendition goes downhill with higher iso, that's why I'm still not so sure if a properly/ettr exposed shot @iso1600 something on the 18mp sensor is always better than iso800 underexposed.

Look at the data.  Bear in mind that the data for ISO 1600 only goes to a luminance of 70, and it's logical to assume that as the color accuracy increases from luminance 30 through 70, it continues to increase through luminance 100.

[Audionut]

Thanks for responding. Exposure seems to be a problem for me since I got my DSLR. Mainly due to a lot of conflicting information i've came across. I would shoot RAW but I only have a 650D so I am limited to H264 for now.

Everyone has an opinion   I would highly recommend that you shot raw photos though.  With non-destructive editing, it's much easier IMO to see the effects of exposure changes.  And if you want to ETTR, you really have no other choice then to shoot raw.

I guess my biggest misunderstanding is should I expose based on luma or RGB? I read somewhere that even if luma channel isn't clipping individual RGB channels clipping can throw away color data.

I would recommend RGB as it's more accurate.  The luma reading sort of mixes all the channels up and gives you an overall exposure reading.  As you read, it can often show no over exposure warning when in fact, one of the RGB channels has indeed clipped.

From your response I assume you suggest ETTR based on readings from the ML histogram rather than the canon EV meter?

ETTR is much much easier to do when you understand the histogram.  It can be done with the Canon meter, but takes some testing to determine where on Canon's exposure meter, overexposure happens.

Exposure is quite a detailed subject.  There are plenty of good resources out there you can read to help increase your understanding (and honestly, they'll probably explain it a lot more accurately and understandable then I can).  You really need to understand it better before attempting to ETTR.  IMO, you will just confuse yourself further.

Some good reading.
http://www.luminous-landscape.com/tutorials/understanding-series/understandexposure.shtml
http://www.luminous-landscape.com/tutorials/understanding-series/understanding-histograms.shtml


Feel free to fire off more questions.  I'll do my best to answer them as clearly and understandable as possible

[newsense]

I guess my only question right now is since I am shooting H264 is my best bet for trying to get a good exposure to use ML histogram set to RGB and try to expose to the center? I will most likely have questions after I follow the links you posted

[Audionut]

IMO yes.  Any editing done to a h.264 stream will be destructive.  So the benefits of ettr can't be exploited.

I would recommend using the Canon exposure meter until you have a good understanding of the histogram.

[AnotherDave]

For the purposes of ETTR in raw video... could a picture profile be created to display a lower exposure (so clients don't freak out when they see everything is "blown out").

Dave