Extreme ETTR example

Started by garry23, January 07, 2020, 10:59:51 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

garry23


Walter Schulz

Boy, this is among the most unnecessary Dual-ISO applications ever ... ;-)
Result looks real good, though!

garry23

Yep, I know  :D

It was an experiment  :)

Having said that, the sun was pretty bright  ;)

Walter Schulz

Yeah, been there, done that.
But may I suggest to add some words for your readers? Some may not be able to detect the nonsense part and may get wrong ideas about when and how to use ML features.

garry23

Hence my use of the word 'extreme' and 'experiment'.

Please feel free to remove the post if you wish.

Cheers

Garry

Walter Schulz

I'm not an administrator and not a moderator. And I see no reason to remove this.

IDA_ML

Garry,

If in doubt on whether you should use Dual ISO or not, do the following:  Carefully expose for the highlights and make sure you do not blow them up.  ETTR helps a lot here.  Now make a test shot and take a look at the histogram.  If the vertical axis on the left cuts through the data then you are clipping the darks and in this case, Dual ISO is your only option.  If, however, the data goes down to zero before it hits the left axis, then leave it as it is - you are well within the dynamic range of your camera and no Dual ISO is necessary.  With your example, this is definitely the case as evident from your histogram data.  If the sun would have come out behind the clouds then contrast would have been much higher and Dual ISO would have been your only option to save the shot.

garry23

Just to be clear thus was an experiment to see if using dual-ISO with ETTR I could get more 'tonal bandwidth' in the image I captured.

My logic was based on an assumption that using dual with ETTR meant that the data on the left, i.e. the 'shadow' areas of the image, was being pushed to the right, thus increasing the tonal resolution, i.e. for post processing. This was not about dynamic range, it was about getting some data to experiment with exploiting with the way digital data is captured, i.e. tonal bandwidth per stop varies in a digital file and is the highest on the right, hence our interest in ETTRing.

Yes doing this I was shooting 'half' the image at ISO 1600, but the 5D3 can handle this, and colour noise can usually be addressed reasonably well in post.

My experiment was not 100% successful, as I should have taken an image without dual on, as a comparison.

So, to be clear, I was experimenting to see if I could increase the tonal 'quality' for post processing, not because I had a high dynamic range.

garry23

I had a few minutes this morning before going to work, so I've added a few words to my original blog post  ;)

Skinny

I thought about it. But without direct comparisons, nothing will be clear... And maybe you need to use less noisy values for the second iso, 800 for example. Since the color resolution in the shadows depends on the noise directly, at least I think so..

Audionut

The tonal quality only really comes into play if you want to stretch the contrast.

The top EV has 2046 levels and thus can be stretched quite aggressively.  When there is only 32 levels, this area of raw captured data can't be stretched as aggressively for the same quality.

If that area of raw capture (32 levels) is in the shadows after PP, it's (edit: likely) a non issue.

In your example I think 100/400 would have been more then enough.

Tonal quality and image noise are semi related but unrelated, if you get my drift.

garry23

@Audionut

QuoteThe tonal quality only really comes into play if you want to stretch the contrast.

...and that was my thinking, as I stretch things out in post. I'm a heavy user of luminosity masks.

In other words, with a scene where you are able to capture all the DR but where the histogram is showing data to the left, with low tonal resolution, would shifting the shadow data by 1-3 stops to the right, using Dual ISO, help processing in post.

I wish I'd had the time to construct a full set of data, but I didn't. I will likely revisit this again and create a better data set of images.

Totally agree with the ISO suggestion, ie use a Dual ISO of, say, 400.


ngemu

would it be better to overexpose to the right, as long as it doesnt clip?

Walter Schulz

?
That's what he did and what ETTR is all about.

Luther

Unless you were running from a lion, the best result would be doing proper HDR. DualISO should be reserved for fast-moving objects, like sports and journalism, IMO. If you have a near static image (such as the example from garry) and have a tripod, do HDR blending instead, IMO.

garry23

@Luther

You are right and I say again, this was a specific (single image) experiment.

Also, exposure bracketing risks introducing artifacts. In this case, although the foreground was mainly static, there was wind and the clouds were moving.

I personally try not to process brackets using HDR algorithms, but prefer to blend brackets with lumonisity masks.

Bottom line: I wish I had kept this experiment to myself  :) ;)

IDA_ML

Quote from: garry23 on January 09, 2020, 07:24:59 AM
Bottom line: I wish I had kept this experiment to myself

Garry,

Most people here, including me, greatly appreciare your work and you know that.  Personally, I am grateful to you for sharing your example with us.  Nobody is trying to attack or criticize you for what you did.  We are just discussing things here and this is how we learn from each other, don't we?

garry23

Feedback always appreciated: hence my addition of  :) ;) at the end of my last post

I guess sarcasm doesn't always translate, nor is it universally recognised as humour  ;D

Audionut

Quote from: garry23 on January 08, 2020, 01:22:20 PM
In other words, with a scene where you are able to capture all the DR but where the histogram is showing data to the left, with low tonal resolution, would shifting the shadow data by 1-3 stops to the right, using Dual ISO, help processing in post.

If you're simply shifting that data (as a whole, so to speak), likely not much at all, since you aren't attempting to take data with a limited number of discreet steps in luminosity and stretch that data to cover a greater range of exposure stops.

You're not taking data captured with 1 or 2 bits of tonal resolution and stretching that data out requiring more display bits.  If you're display chain is not accurate and thus shrinks areas of luminosity and expands other areas of luminosity, that's another issue.

If the data of interest is way over to the left (ETTL) and contains tonal resolution issues in the source data (likely hidden by noise (think dithering)) and you clean the noise and then shift that data to the right......

If you want to test (and we would all appreciate it), I wouldn't bother trying to capture good looking display images.  I would focus on capturing good data just in the shadows and focus on post processing that shadow data (highlights be damned).

I agree with your assessment of temporal artifacts in traditional HDR capture and processing, although maybe I just suck at it.  And mistakes make for the best discussion when they don't descend in to pettiness.  ;D

Skinny

Guys, if I understand correctly, the whole point is, for example, we need to raise the shadows a little bit. But they are not clipped or something.

And we have two options, to increase them digitally, or to use higher iso for the shadows. The second method is analog, and it doesn't rely on the A/D converter resolution.
Another thing is that the dynamic range of the camera is more than enough to apply a little digital gain in the shadows...

I don't think that we will see any difference in the noise. But it can be in the tonal resolution.
The correct test would be to shoot a smooth gradient in the shadows and try to make it brighter.. Who knows..

sahrenity