Pages: 1 ... 5 6 [7] 8 9   Go Down

Author Topic: ETTR vs ISO  (Read 71382 times)

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

In low light photography where you cannot give the sensor enough exposure to get close to blowing any highlights through full wells, I can only see a case for erring a bit on the low side with analog gain (ISO speed setting) to avoid clipping of highlights by amplifying them beyond the maximum ADC output level. I cannot see any benefit beyond a bit of a safety margin, and the idea of under-amplifying so much that three or more stops of digital amplification are needed in post-processing sounds unwise --- more so if many widely-used raw convertors will then introduce a color shift when you push the levels up by three or more stops.

One thing to beware of is the blind pursuit of higher engineering DR values, like the talk of increasing the engineering DR by one stop for each stop of lowered ISO amplification: beyond some point, that extra DR is just unused numerical levels above any level present in the signal. If one ISO level avoids clipping any highlights in any color channel, then each subsequent halving of the ISO gain adds an extra zero leading bit before the first significant bit, which adds a stop of engineering DR, measured from the maximum ADC output level down, but adds nothing to signal quality, and moves you closer to increased shadow noise due to the increased ADC quantization noise.

With the metering of most modern system cameras producing raw files in which the metered level is placed far enough below the maximum level that there is almost always a significant margin of unused levels near the top, under-amplifying is in most cases going to offer no IQ benefit, while adding avoidable complications to the raw conversion process --- especially if you then need to construct custom profiles to avoid color shifts.
Logged

jrsforums

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1288

In low light photography where you cannot give the sensor enough exposure to get close to blowing any highlights through full wells, I can only see a case for erring a bit on the low side with analog gain (ISO speed setting) to avoid clipping of highlights by amplifying them beyond the maximum ADC output level. I cannot see any benefit beyond a bit of a safety margin, and the idea of under-amplifying so much that three or more stops of digital amplification are needed in post-processing sounds unwise --- more so if many widely-used raw convertors will then introduce a color shift when you push the levels up by three or more stops.

One thing to beware of is the blind pursuit of higher engineering DR values, like the talk of increasing the engineering DR by one stop for each stop of lowered ISO amplification: beyond some point, that extra DR is just unused numerical levels above any level present in the signal. If one ISO level avoids clipping any highlights in any color channel, then each subsequent halving of the ISO gain adds an extra zero leading bit before the first significant bit, which adds a stop of engineering DR, measured from the maximum ADC output level down, but adds nothing to signal quality, and moves you closer to increased shadow noise due to the increased ADC quantization noise.

With the metering of most modern system cameras producing raw files in which the metered level is placed far enough below the maximum level that there is almost always a significant margin of unused levels near the top, under-amplifying is in most cases going to offer no IQ benefit, while adding avoidable complications to the raw conversion process --- especially if you then need to construct custom profiles to avoid color shifts.

Thanks. BJL.

Took me a couple readings, but I think to old brain finally got it.

In Emil's statement (which I posted very early in this thread), he was making the point that, with ISOless sensors, it bought nothing to raise ISO to set the histogram further to the right, ETTR.  What you are saying, I think, is that doing so does not cost you anything either....the "increased" DR has no value as it is highlights of no value.

...and, with sensors such as the Canons, increasing the ISO could get you less noise in the shadow areas.

John
Logged
John

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog

One thing to beware of is the blind pursuit of higher engineering DR values, like the talk of increasing the engineering DR by one stop for each stop of lowered ISO amplification: beyond some point, that extra DR is just unused numerical levels above any level present in the signal.

Yes Bill, although I fail to see how that would be undesirable.  Once you have maxed out Exposure (that is the longest shutter speed and the lowest f/number that your technical/artistic constrainsts allow), raise ISO from base to give you the best SNR possible without blowing desirable highlights and stop there.  This strategy allows you to capture all of the relevant information from the scene.  It makes zero difference to IQ whether the last few bits are unused or the first few bits are filled with random noise.  What matters is that you got all of the information you wanted.  In fact with this strategy not only do you get all of it, you need to worry about one less variable (ISO and blowing highlights) and you get smaller files to boot.  What's there not to like?

As for twists, no point blowing things out of proportion.  Here is a capture by a D800e at base ISO that the guy says requires a 6 stop brightness push for pleasing tones (don't try this with Canon, though): I don't know that I'd go that far, but even then it's nice to know you can.  Click through to see whether you can spot the twist in LR: I can barely see it.

Cheers,
Jack
« Last Edit: May 27, 2013, 05:23:04 pm by Jack Hogan »
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word

Once you have maxed out Exposure (that is the longest shutter speed and the lowest f/number that your technical/artistic constraints allow), raise ISO from base to give you the best SNR possible without blowing desirable highlights and stop there.  This strategy allows you to capture all of the relevant information from the scene.  It makes zero difference to IQ whether the last few bits are unused or the first few bits are filled with random noise.  What matters is that you got all of the information you wanted.  In fact with this strategy not only do you get all of it, you need to worry about one less variable (ISO and blowing highlights) and you get smaller files to boot. 

Once you have given up on ETTR because you don't have enough light to make an image with the f-stop and shutter speed you want at base ISO, and you start letting the histogram slide to the left, something magical happens: you no longer have to worry about exposure. The light changes; you ignore it, knowing it will make no difference to you. You don't have to continue to consult the histogram; you don't care what it looks like, as long as it hasn't crept alarmingly leftward, and you'd know if that were happening. It's remarkably freeing. You work faster, and you are more able to concentrate on your subject.

Jim

RFPhotography

  • Guest
Re: ETTR vs ISO
« Reply #124 on: May 27, 2013, 07:10:50 pm »

Bob, I am no expert on unity gain, but I believe that Jim proved a few threads ago that Unity Gain is in practice only meaningful in very specific applications such as when one does heavy duty stacking for astrophotography.  As far as I understand today PDR Shadow Improvement is as good as it gets in helping to make the proper SNR/DR compromise.

Jack

Thanks, Jack and Jim.  I'd think that, like PDR/SI, it would be useful in any situation where you were having to increase ISO to get a 'normal' exposure (or even an ETTR'd exposure in cases where ETTR above base ISO is useful).  Knowing what point to stop bumping the ISO to prevent digital multiplication and potential blowing of highlights would be good, no?  That assumes that one doesn't need a 'normal' looking exposure and has the time to make adjustments after the fact.  This article seems to make a reasonable point in that regard, http://www.ishootshows.com/2009/01/28/push-processing-and-unity-gain/.

Based on the small sampling in that article and comparing to the information on Claff's website, it does appear that the two are quite close, as Jim surmises.
« Last Edit: May 27, 2013, 07:20:57 pm by BobFisher »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

Once you have given up on ETTR because you don't have enough light to make an image with the f-stop and shutter speed you want at base ISO ... you no longer have to worry about exposure. The light changes; you ignore it, knowing it will make no difference to you ...
Jim, yes that approach makes sense in low and variable light where the traditional approach of "on meter exposure" would involve jumping around the ISO speed setting: just use a value that is sure not to be above the low end of the range, so that some shots might be "under-amplified" (underexposed with default JPEG conversion) but none are over-amplified to the point of highlight clipping.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

... I fail to see how that would be undesirable. 
To answer, I will quote my post that you are are quoting:
"adding avoidable complications to the raw conversion process" like having to adjust the levels individually for each conversion whereas a high proportion of on-meter exposures do not need this, and "especially if you then need to construct custom profiles to avoid color shifts." So only inconvenience, and Jim Kasson has just pointed out one case where the extra PP effort can be well worth it, but when you can easily enough use the ISO that "fits" with your choice of shutter speed, aperture and the light levels, why make work for yourself with post-processing manipulations and run at least a risk of color twist at more extreme levels under-amplification? My man warning is that some people are claiming that this extra effort gives an IQ advantage that does not in fact exist in this particular situation.
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: ETTR vs ISO
« Reply #127 on: May 27, 2013, 09:20:50 pm »

I've finished my Canon 5DIII simulation, and I think that the advantages of increasing the ISO that people are seeing can be explained by the post amplification portion of the read noise, which I put at 7.5 analog-to-digital converter (ADC) counts (aka ADUs). The effect of this relatively large noise can be countered by increasing the gain of the amplifier ahead of the ADC, which is what you do when you turn up the ISO.

Here are the photon corrected signal-to-noise ratio (SNR) results for the simulated 5DIII when presented with light levels that produce constant mean ADC counts a factor of two (one stop) apart from near-saturation to near the noise floor. I did not simulate any digital gain; I assumed it is all analog. The results are for one of the (assumed identical) green channels only.



You can see that increasing the ISO helps all the way to ISO 12800, especially in the darkest tones. The reason for the bump from ISO 100 to ISO 400 in the lightest tones is that turning up the ISO lessens the effect of pixel-response non-uniformity (PRNU). I don't consider this to be an effect worth worrying about in cameras with PRNUs below half a percent.

Compare the Canon results with the D800E ones:



The simulated D800E is essentially "ISO-less" ( I don't like that term, but it has acquired a certain currency). The real D800E has more problems than the simulated one at ISOs above 1600-3200, probably because of the digital gain. Note that the Canon SNRs are slightly better than the Nikon ones for brighter tones at the higher ISOs; that's because the Nikon's full well capacity is lower. If the Nikon images were res'd down to Canon levels, this advantage would disappear.

Jim
« Last Edit: May 27, 2013, 09:40:53 pm by Jim Kasson »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
noise sources after analog amplification, before ADC
« Reply #128 on: May 27, 2013, 09:35:02 pm »

I've finished my Canon 5DIII simulation, and I think that the advantages of increasing the ISO that people are seeing can be explained by the post amplifications portion of the read noise ...
This fits with what I have gathered from other sources: Canon's recent sensors are unusual in having early analog amplification (on-chip) but late ADC (off-chip) and so a significant opportunity for noise to enter the signal between analog amplification and ADC. In contrast, CCD's and maybe older Canon CMOS sensors did analog amplification later, off-chip just before ADC, and most other modern CMOS sensors do both analog amplification and ADC very early, at the bottom of each column of pixels. In either case there is far less opportunity for noise to enter between analog amplification and ADC. (But the ADC itself can add some noise, quantization noise in particular.)
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: noise sources after analog amplification, before ADC
« Reply #129 on: May 27, 2013, 09:43:44 pm »

Canon's recent sensors are unusual in having early analog amplification (on-chip) but late ADC (off-chip) and so a significant opportunity for noise to enter the signal between analog amplification and ADC. In contrast, CCD's and maybe older Canon CMOS sensors did analog amplification later, off-chip just before ADC, and most other modern CMOS sensors do both analog amplification and ADC very early, at the bottom of each column of pixels. In either case there is far less opportunity for noise to enter between analog amplification and ADC. (But the ADC itself can add some noise, quantization noise in particular.)

That makes sense. Thanks.

Jim

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog

To answer, I will quote my post that you are are quoting:
"adding avoidable complications to the raw conversion process" like having to adjust the levels individually for each conversion whereas a high proportion of on-meter exposures do not need this, and "especially if you then need to construct custom profiles to avoid color shifts." So only inconvenience, and Jim Kasson has just pointed out one case where the extra PP effort can be well worth it, but when you can easily enough use the ISO that "fits" with your choice of shutter speed, aperture and the light levels, why make work for yourself with post-processing manipulations and run at least a risk of color twist at more extreme levels under-amplification? My man warning is that some people are claiming that this extra effort gives an IQ advantage that does not in fact exist in this particular situation.

Yes, potential brightness correction in post is a compromise with the potential extra DR, depending on your workflow.  I shoot Raw+Jpeg and fine tune all of my keepers, so that's not an issue for me, but I can see how it could be for others.

I think the main take-away from this approach is that it teaches photographers to decouple Exposure (photon counting, shutter speed, f/number) from ISO (processing, perceived brightness, SNR, DR).  The remaining problem to solve (the one that is causing a lot of confusion) is that only one parameter, ISO, is used to control both objective IQ related quantities like SNR/DR and subjective ones like perceived output image brightness.  It's my guess (and wish) that in a generation or so what we call ISO in-camera will be split into two components, initially hidden from the beginner: a 'gain' to control how the sensor output gets processed analogically before being written to the Raw data and a 'Perceived Lightness' target brightness for the OOC image independent of the average values of the Raw data.

Cheers,
Jack
« Last Edit: May 28, 2013, 03:25:04 am by Jack Hogan »
Logged

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 798
    • Hikes -more than strolls- with my dog
Re: ETTR vs ISO
« Reply #131 on: May 28, 2013, 03:12:17 am »

I've finished my Canon 5DIII simulation, and I think that the advantages of increasing the ISO that people are seeing can be explained by the post amplification portion of the read noise, which I put at 7.5 analog-to-digital converter (ADC) counts (aka ADUs). The effect of this relatively large noise can be countered by increasing the gain of the amplifier ahead of the ADC, which is what you do when you turn up the ISO.

Good show, thanks Jim.  Makes one wonder whether to save space one should stick to 12-bit mode.
« Last Edit: May 28, 2013, 04:04:29 am by Jack Hogan »
Logged

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro

I think the main take-away from this approach is that it teaches photographers to decouple Exposure (photon counting, shutter speed, f/number) from ISO (processing, perceived brightness, SNR, DR).  The remaining problem to solve (the one that is causing a lot of confusion) is that only one parameter, ISO, is used to control both objective IQ related quantities like SNR/DR and subjective ones like perceived output image brightness.  It's my guess (and wish) that in a generation or so what we call ISO in-camera will be split into two components, initially hidden from the beginner: a 'gain' to control how the sensor output gets processed analogically before being written to the Raw data and a 'Perceived Lightness' target brightness for the OOC image independent of the average values of the Raw data.

Cheers,
Jack

What I would like is a live view mode or EVF with two sets of "Blinkies" , one for highlights and the other for the shadows, based on actual raw values, not rendered values:

- The highlight blinkies just show saturated or blown raw values.
- The shadow blinkies are a little more complex, the idea is to be able to configure a minumun acceptable SNR ratio (for non-technicals it could be just a "quality" factor) so that any areas with a raw level that will result in lower than desired SNR will blink.

With those two blinkies, you could just play with the 'gain' until you achieve the best compromise between clipped highlight and noisy shadows. This of course for those that are confortable with RAW processing.

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051

What I would like is a live view mode or EVF with two sets of "Blinkies" , one for highlights and the other for the shadows, based on actual raw values, not rendered values:

- The highlight blinkies just show saturated or blown raw values.
- The shadow blinkies are a little more complex, the idea is to be able to configure a minumun acceptable SNR ratio (for non-technicals it could be just a "quality" factor) so that any areas with a raw level that will result in lower than desired SNR will blink.

With those two blinkies, you could just play with the 'gain' until you achieve the best compromise between clipped highlight and noisy shadows. This of course for those that are confortable with RAW processing.

I believe that Magic Lantern (3rd party software for Canon DSLRs) provide raw-based histogram, blinkies etc.

http://www.magiclantern.fm/whats-new/95-tutorial/catexposure/125-new-fast-zebras

-h
Logged

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro

I believe that Magic Lantern (3rd party software for Canon DSLRs) provide raw-based histogram, blinkies etc.

http://www.magiclantern.fm/whats-new/95-tutorial/catexposure/125-new-fast-zebras

-h

It looks nice, but it seems to me it works with rendered data and not raw values. It is not clear to me if the "shadows" warning just show values (again rendered) near zero or if you could customize a SNR value.

Thanks, regards.

Edit: I found more info and you can setup the shadow threshold but I'm pretty sure it is based on rendered values.
« Last Edit: May 28, 2013, 05:22:35 am by FranciscoDisilvestro »
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: ETTR vs ISO
« Reply #135 on: May 28, 2013, 06:59:42 am »

http://www.magiclantern.fm/forum/index.php?topic=5149.msg31247#msg31247
Quote
Alright, now you have true RAW histogram and zebras. 5D2 only.

To use, simply choose RAW in Canon menu and set histo/zebra to RGB. No new menus - it should just work.

https://bitbucket.org/hudson/magic-lantern/commits/115f297ac7c8
Quote
After some quick tests, RAW histogram and the zebras are matching exactly what I see from RawDigger

Sweet!
Quote
Black level got fixed... working on many digic IV cameras... you just have to find the bins.
Seems that it is sort of an early feature, but anyways...

Strange that it takes reverse-engineering to give us the features that we want, while the camera manufacturers seems to focus on in-camera HDR, the infamous "PictBridge" button and other features that seems to appeal more to casual users (catered for by iPhones, superzoom and m4/3?) than those interested in clunky, expensive 35mm SLR cameras.

-h
« Last Edit: May 28, 2013, 08:42:00 am by hjulenissen »
Logged

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro
Re: ETTR vs ISO
« Reply #136 on: May 28, 2013, 07:14:47 am »

Great!, It looks promising, thanks for sharing.

I only wonder how much powerful and useful options we could have if only the camera makers offered a documented SDK to customize the camera firwmare.

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: ETTR vs ISO
« Reply #137 on: May 28, 2013, 10:19:08 am »

Good show, thanks Jim.  Makes one wonder whether to save space one should stick to 12-bit mode.

I'll run a test.

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600

Yes, potential brightness correction in post is a compromise with the potential extra DR, depending on your workflow.  I shoot Raw+Jpeg and fine tune all of my keepers, so that's not an issue for me, but I can see how it could be for others.
I also shoot raw+JPEG, and I still see one inconvenience to substantially "under-amplifying" in low light --- along with a potential solution, and one that is indeed already more or less offered by some cameras.

The problem is that if I used an ISO sensitivity setting that for some shots is well below what would be needed for "on-meter" exposure, the JPEG's and previews of the raw files will sometimes be way too dark, and then the problems start before I get to fine-tuning my keepers:
it becomes necessary to make rough level adjustments on every "under-amplified" image just to get a preview that allows me to decide which shots _are_ the keepers.

One solution for me involves:
a) A preview image (the OOC JPEG, or default raw->JPEG conversion) that has been amplified to roughly correct levels.
b) Raw files in low-light situations that have lower levels from "conservative amplification": that is, the raw level histogram has a significant gap at right.
Camera makers using almost "ISO-less" sensors could accommodate both goals by keeping the analog gain lowish, thus allowing an abundant safety margin against highlight clipping by over amplification in the raw file, but tagging the raw file with information about the intended exposure level, with this used in the in-camera JPEG conversion and in default conversion of the raw files by any software that knows about the camera (which includes Adobe DNG Convertor, Lightroom etc. once they are updated for each new model of camera.)

The good news is that many camera makers already do something like this, so I think we might be over-thinking a situation that has already been addressed. I refer to cameras that use conservative analog gain at elevated ISO speed settings along with providing information about this that is used by raw convertors and in making in-camera JPEGs, thus offering both raw files with abundant highlight headroom and default JPEG versions with appropriate "viewable" brightness levels. These camera models are fairly easy to detect: they are the ones that offer substantial blown-highlight recovery from raw files, and for which the DxO measurement of "sensitivity" at elevated ISO settings is significantly less than the ISO sensitivity setting on the camera. That is, those cameras which some misguided followers of the over-broad ETTR doctrine ("the raw histogram should always be as far to the right as possible", even in low light/high ISO speed situations where read noise overwhelms ADC quantization noise) accuse of "cheating on" or "overstating" their ISO speed settings.
« Last Edit: May 28, 2013, 10:39:57 am by BJL »
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: ETTR vs ISO
« Reply #139 on: May 28, 2013, 10:48:13 am »

Makes one wonder whether to save space one should stick to 12-bit mode.

Here are the simulation results, with the curves being the 14-bit SNRs less the 12-bit SNRs:



I don't know quite what to make of it. It shows that 14-bits gives an improvement in SNR for shadows at low ISO, but actually hurts a tiny bit for shadows at high ISO. In the middle tones and above, it makes no difference.

Jim
« Last Edit: May 28, 2013, 10:50:31 am by Jim Kasson »
Logged
Pages: 1 ... 5 6 [7] 8 9   Go Up