[mccririck]

Exactly, the art of producing a decent image would be very difficult to codify. And codifying would reduce photography from an art form to series of same looking images. There are some things you never want to give up control of, and decisions about how to interpret the light you are seeing with camera is the most obvious. You do it by knowing your tool, and experience. Which is why this argument is futile. Every shortcut has it's issues ETTR is not different. That being said there will be scenes where ETTR is appropriate. It's the opportunities you miss when it's not that will be the problem.

If I had to make up an acronym for what I do it would be "Expose for the Available Light". With modern high DR cameras, expose to the left will be as appropriate in as many situations as expose too the right will be. And soemtimes "keep the highest part of the curve in the centre (or to the left or right of centre)" is the appropriate approach. I fail to see how ETTR dude came to the conclusion that adopting 1 of of those possibilities is the correct one. Possibly a function of your average colour film having half the DR of a K-1 has something to do with it. No shadow detail could be rescued from unexposed areas of film, there was nothing there.

Bottom line, if you're learning photography there are no acceptable cheats. You have to chimp, look at your histogram, then decide how to adjust accordingly. I haven't come up with any rules to help with that. IMHO, each case stands on it's own.

I feel you are ignoring the fact that far more detail is saved the further to the right you go. On the left much less detail is recorded and you cannot get it back after you've taken the shot. But you can adjust in post to get the shadows back (with detail preserved).

That is the whole point of ETTR.

[normhead]

I feel you are ignoring the fact that far more detail is saved the further to the right you go. On the left much less detail is recorded and you cannot get it back after you've taken the shot. But you can adjust in post to get the shadows back (with detail preserved).

That is the whole point of ETTR.

You can't get back detail whether you blow out the right or left, that doesn't make any sense. You get 13 EV of resolution no matter where you set the centre point, if you're shooting a 20 EV scene and you select left or right depending on where you want to select your EV, you don't get extra for ETTR. You get the same 13 EV, just a different 13 EV. Imagine a scale 20 CM long with a sliding bar 14 cm long. You can move your 14 EV to the left of right within the 20 cm bar, but you never get the whole 20 cm. If each cm represents 1 EV, you have to realize each time you move the bar, say you're in the middle with a 14 CM(EV) bar. There will be three CM to either side. now you can shift the bar to the left so you have 2 cm to the left, and 4 to the right. That will change how the image will look, but it doesn't get you more of anything. If you've ever had to match a grey scale with black and white film, you'll understand exactly what I'm talking about.

By the way you continue to say detail when you mean colour depth. You seem to be confusing two different values. Usually detail is associated with resolution, not colour depth. MTF test charts measuring detail usually are black and white.

The disadvantage to ETTR, in both film and digital is if you move your 14 CM (EV) bar, all the way to the right, the image will appear to be flat and much harder to achieve decent contrast. That coupled with the fact that an 8 bit jpeg can't reproduce all that detail anyway, which is a good thing because your eye can only discern about 16 million gradations, roughly jpeg output.).

So between the fact that ETTR only makes any sense when the DR of the scene is less than the EV that the camera is capable fo capturing, even then, you may want to keep you exposure to the left for contrast. You only need what your eye can see, and having more colour depth than your eye can see make no sense at all to the final image. Having extra colour depth is very rarely an issue as even right on the left edge is hundreds if not thousands more colour depth than you can output in a jpeg. By keeping your curve as close as possible to the left, you maximize your out put. Move the curve at all to the right and you blow out more highlights. For in-between shots where the EV of the camera exceeds that of the scene captured, you'll find keeping the curve to the left requires less post processing, because the numerical values more closely resemble your jpeg output.

[stevebrot]

how are you supposed to get your exposure right?

Trust the matrix metering, turn the blinkies off, evaluate the scene in advance and if in doubt, do a two stop bracket with -1 EC.

That being said...possibly reconsider the notion of getting one's exposure "right". ETTR is a tool, not a religion. For difficult lighting situations, there are three fallback strategies/tools in my kit, each of which is definitely old school and all are are guaranteed to work within the limitations of the sensor (or film):

• Measure light incident to the subject using either a gray card or handheld meter; either will result in a centered histogram with preservation of shadow detail.
• Handheld meter or spot metering to measure a critical portion of the subject to allow placement of exposure to that segment of the scene
• Bracket with the intent to merge in post if needed

It is quite possible that neither of the first two will prevent clipping on one end or the other,* but taking the time to either place the histogram or place the critical values should allow one to determine the potential need for HDR, the only option short of simply letting the highlights or shadows to clip.


Steve

* After all, one has only 14 stops (two of which are potentially clipped) to work with on the RAW capture and only eight if publishing to JPEG.

[photoptimist]

I feel you are ignoring the fact that far more detail is saved the further to the right you go. On the left much less detail is recorded and you cannot get it back after you've taken the shot. But you can adjust in post to get the shadows back (with detail preserved).

That is the whole point of ETTR.

Norm's not ignoring it so much as pointing out the trade-offs -- improving what's on the left of the histogram by moving it to the right may mean sacrificing what's on the right the histogram.

The other interesting issue is that moving the histogram to the right to improve the quality fo the details in the shadows may force a loss of details in other ways. The only way to move the histogram to the right is to collect more light which means either slowing the shutter speed (and possibly losing detail to motion blur) or opening the aperture (and losing detail to shallower DoF and possibly softer lens resolution). Boosting the ISO does not help in the case of modern ISO-invariant sensors such as the K-1.

ETTR is a useful concept but it comes with some trade-offs and issues that make it less than universally useful.

[mccririck]

Colour depth is part if detail, surely? And you cannot get that back if you expose to the left. And ETTR is not aboutblowing highlights - obviously you cannot get that back.

[normhead]

Colour depth is part if detail, surely? And you cannot get that back if you expose to the left. And ETTR is not aboutblowing highlights - obviously you cannot get that back.

Contrast has more effect on the perception of detail than straight colour depth.

Having extra colour depth is very rarely an issue as even right on the left edge is hundreds if not thousands more colour depth than you can output in a jpeg

.

Already addressed. Having 1000 shades of grey instead of 20,000 makes no difference to the final image. They are both far more than you can see. You can't get it back, but, cameras are designed as they are, because you don't need it. So, no ridiculously over sampled colour depth is not part of detail. You can't tell the difference in most cases between 16 black values and 256 black values. And no colour depth isn't that important in detail. Contrast and micro contrast and lens resolution are much more important.

You seriously need to try some photos to test these things out. Those of us who regularly bracket deal with this issue day in day out. It's pointless having a theoretical discussion about it. Honestly, take a nice bright scene and bracket, then examine what you get with the various exposure settings. What I'm discussing will become immediately apparent. It's not rocket science. And with digital, bracketing is cheap.

[photoptimist]

Contrast has more effect on the perception of detail than straight colour depth.

.

Already addressed. Having 1000 shades of grey instead of 20,000 makes no difference to the final image. They are both far more than you can see. You can't get it back, but, cameras are designed as they are, because you don't need it. So, no ridiculously over sampled colour depth is not part of detail. You can't tell the difference in most cases between 16 black values and 256 black values. And no colour depth isn't that important in detail. Contrast and micro contrast and lens resolution are much more important.

This is not true -- DR is essential for resolving detail. As DR diminishes, the amplitude of the noise approaches the amplitude of the signal and then the ability to resolve the signal diminishes. The effect is very obvious in high-ISO shots and shots with heavily-lifted shadows -- the luminance and chrominance noise interferes with seeing the subject. And if you take an 8-bit image and convert it to 4-bit, then all of the shadow detail that was in 8-bit grays values of 0 to 15 becomes just an unresolvable black 0.

Sure, DR isn't sufficient. Lots of other issues can also reduce detail: subject or camera motion, too slow a shutter, too large an aperture, too small an aperture, missed focus, a crappy lens, heat haze, etc. (Note: there's also an interesting interplay between DR and the ability to correct for other optical causes of blur -- a low-ISO image can withstand a lot more sharpening than can a high-ISO image.)

You seriously need to try some photos to test these things out. Those of us who regularly bracket deal with this issue day in day out. It's pointless having a theoretical discussion about it. Honestly, take a nice bright scene and bracket, then examine what you get with the various exposure settings. What I'm discussing will become immediately apparent. It's not rocket science. And with digital, bracketing is cheap.

Bracketing is cheap but I wonder how often the image picked in post ends up being the ETTR image that has pulled the shadows to the right but not clipped anything. Bracketing is good risk management strategy, but it should also be a learning tool especially for photographers of moving subjects and dynamic scenes where there is no time to bracket.

[stevebrot]

And ETTR is not aboutblowing highlights - obviously you cannot get that back.

You can't get clipped shadows back either, regardless of what clever dithering by software may trick us into thinking. The paucity of low value data is very real and perhaps more important to image quality than S/N ratio.


Steve

[stevebrot]

Bracketing is cheap but I wonder how often the image picked in post ends up being the ETTR image that has pulled the shadows to the right but not clipped anything.

I have spent a fair amount of time, as of late, analyzing the initial import behavior of Lightroom for images that press the envelope in regards to tonal values, specifically using synthetic input files. There is a fair amount of unexpected reshuffling and value reassignment to normalize both tails of the histogram.


Steve

[normhead]

This is not true -- DR is essential for resolving detail. As DR diminishes, the amplitude of the noise approaches the amplitude of the signal and then the ability to resolve the signal diminishes.

You aren't even talking about what I was talking about. I was discussing colour depth, not Dynamic range, I think you must have read it too quickly. But with regards to that, bringing dynamic range and noise into the equation obscures the issue. You avoid loss of DR by shooting 100 or 200 ISO. That has nothing to do with either colour depth or ETTR.

---------- Post added 07-12-19 at 10:37 AM ----------

I have spent a fair amount of time, as of late, analyzing the initial import behavior of Lightroom for images that press the envelope in regards to tonal values, specifically using synthetic input files. There is a fair amount of unexpected reshuffling and value reassignment to normalize both tails of the histogram.


Steve

Thats why we go to 14 bit Raw. More colour depth leads to more latitude in that reshuffling process and more realistic looking colour after a shuffle, in both the high end and low end. But in the end you are still going to squeeze a 14 bit raw into an 8 bit jpeg (so you can actually see some output) and toss most of that information that was preserved in the 14 bit raw.

The issue here is not that colour depth is of no use. It's that a 14 bit raw has 6 bits more than you need for the output. The rest is work space. That's a lot of leeway. The suggestion seems to have been that if you don't expose to the right you won't have enough to work with. That's simply not true. That extra 6 bits provides you a ton of work room even at the narrow (black) end of the spectrum.

[stevebrot]

That extra 6 bits provides you a ton of work room even at the narrow (black) end of the spectrum.

Sadly, the headroom is all in the high values, at least from the purely data perspective. Low values have the same stop-wise representation in the RAW data regardless of bit depth (0, 1, 2, 4, 8, etc. with one being the lowest detectable value and zero being indeterminate). Extra bit depth gives our tools more latitude, but only with compromise at the low end. When one dithers a faint twig, it is the twig that is lost.


Steve

[normhead]

Sadly, the headroom is all in the high values, at least from the purely data perspective

Ya, I really don't think much about a raw data perspective, my viewpoint is all empirical. What can I do with it?
My point to date in terms of raw data, is, using 14 bit raw you never max it out. You always have 100 times more than you can use in a jpeg.
The reason for going to 14 bit raw is to avoid the limitations of 8 bit jpeg, it accomplished what it set out to do, and eliminated the need to worry about colour accuracy and lack of gradation in the low end. 8 bit is fine for what you see. 14 bit is what you need to shift values in post. But I've seen no evidence that you need more than 16 different blacks in your final image. As has been pointed out repeatedly, we regularly pull detail out of completely black values in the original jpeg thumbnails. Maybe you had to worry about such things with film, not so much with digital. There is a lot of detail down there you will never want to access, it's there just in case, and usually the case involves you totally blew the exposure by shifting to the right so much you left part of the image unexposed because you didn't exceed minimum sensitivity. 14 bit raw is serious overkill situation. Even 12 bit is.

If your histogram tells you you're almost all the way to the left, you're good to go, in most circumstances. And I'd argue that's the best approach. But the other perspective is you want the mid section of the course on your subject. Based on what I see, the theoretical stuff is just lame attempts to explain my results. Honestly, I don't know and don't care. It's amazing how often I get drawn into these theoretical discussions with folks disputing my practice and results based on theoretical nit picking. I use theory, only to explain my results. It's not like I and others don't post relevant images.

It's completely inappropriate to try and dispute practical knowledge with theoretical objections. Show me what you mean with practical examples and we can talk. I'm sharing practical observations of how it works, trying to provide some theoretical background. But it's based on my results. Theoretical objections may be of interest for discussion, but they can't change the observed results.

Without practical examples I can't tell if folks are just blowing hot air or are actually on to something. Repeatable samples are critical to extending knowledge. I don't really want to know what you think. I want to know what benefit to me there might be to thinking that way. Then I still want to evaluate if your thinking is straight or you just lucked into something that is actually unrelated to your logic.

My current practice is "Get the back screen jpeg and histogram as close as you can to what you want, clean it up in post processing." There is all the latitude you need with that approach in even a 12 bit Raw.

IN the case of ETTR, I thought about it, I tried is, it doesn't stack up. It will cost you images. Chimping and using the histogram is better, in my experience. The histogram may not be precise, but it's consistent, but you can learn to interpret what you see. I see ETTR as a theory relevant to the days before chimping was possible.

If someone tells me they do it that way... no problem, believe what you want. If they are implying it's what everyone should use, that's a problem.

[stevebrot]

Ya, I really don't think much about a raw data perspective, my viewpoint is all empirical. What can I do with it?
My point to date in terms of raw data, is, using 14 bit raw you never max it out. You always have 100 times more than you can use in a jpeg.
The reason for going to 14 bit raw is to avoid the limitations of 8 bit jpeg, it accomplished what it set out to do, and eliminated the need to worry about colour accuracy and lack of gradation in the low end. 8 bit is fine for what you see. 14 bit is what you need to shift values in post. But I've seen no evidence that you need more than 16 different blacks in your final image.

I agree completely regarding the practical aspects, more bits is obviously better. Such is quite obvious in PP. As for the 16 different blacks, don't worry, they aren't there.

As has been pointed out repeatedly, we regularly pull detail out of completely black values in the original jpeg thumbnails.

I am unaware of those claims, but examples on this site have been unimpressive. To be fair, my own efforts suffer the same, which is one reason why I seldom do shadow pulls.

Maybe you had to worry about such things with film, not so much with digital.

I still do both and that has not been my experience, at least not on the negative. After scanning to 16 or 32-bit TIFF, the usual digital constraints kick in, however.

There is a lot of detail down there you will never want access, it's there just in case, and usually the case involves you totally blew the exposure by shifting to the right so much you left part of the image unexposed because you didn't exceed minimum sensitivity.

You hit the nail right on the head. There is a huge risk of simply recording nothing but noise in the shadows if one tries too hard to avoid highlight clipping. As for the amount of detail available, there are only eight shades of gray (pixel-wise) in the bottom four stops, so best to not expect much there.*


Steve

* Gross over-simplification, but still the truth in terms of capture data.

[normhead]

You hit the nail right on the head. There is a huge risk of simply recording nothing but noise in the shadows if one tries too hard to avoid highlight clipping. As for the amount of detail available, there are only eight shades of gray (pixel-wise) in the bottom four stops, so best to not expect much there.*

8 shades of grey in the bottom 4 stops?
What is that in reference to?

That would make the K-1 practically a 10 EV camera, with no detail available in the bottom 4 stops. But it would explain why my images sometimes look better when I adjust my curve so the curve goes off the left side of the histogram and is centred to the left, as opposed to the right. That would essentially make the K-1 a 10 EV camera with a lot more available shadow detail, but, I still need a reference of some kind. I'm not sure it's that extreme.

After all, I decide where my black point is going to go, not the camera. If I choose to put 8 bits behind the black point I can (and that would give me 16 million gradations of black) and that would correspond to the bottom 4 stops in output and I'd still have 6 bits left for the right side of the curve. It's not arbitrary.

[stevebrot]

I still need a reference of some kind. I'm not sure it's that extreme.

I will PM you as the explanation skews off-topic.


Steve