A7r and ETTR

[JimVehe]

Has anyone figured out a good manual exposure protocol.  The zebra blinking only seems to show under exposure and an ETTR histogram has a good stop and a half more room when opened in Lightroom or ACR.  Even chimping with the blinkies under exposes by more than a stop.  I have had limited time to shot with the camera over the past two months.  When you get the exposure right the image quality is quite remarkable and an amazing amount of dynamic range.  I am hoping that I am just missing something rather than having to learn from experience how much more to over expose form the indicated histogram.

[Fine_Art]

As usual the histogram is based on the in camera jpg. Turn the image contrast down. Use portrait rather than landscape. The camera jpg will better match the RAW capability.

[JimVehe]

I will play with that possibility this week.  I had been shooting RAW only.  I will try RAW plus jpeg in various styles as well as changing the settings for jpeg and still shoot Raw only.  I was hoping that a live view camera would decrease the need to chimp.  On My last digital Leica M the only thing that the screen was good for was an exposure chimp, but it was pretty close to clipping if blinking.

Thanks

[Jim Kasson]

Has anyone figured out a good manual exposure protocol.  The zebra blinking only seems to show under exposure and an ETTR histogram has a good stop and a half more room when opened in Lightroom or ACR.  Even chimping with the blinkies under exposes by more than a stop.  I have had limited time to shot with the camera over the past two months.  When you get the exposure right the image quality is quite remarkable and an amazing amount of dynamic range.  I am hoping that I am just missing something rather than having to learn from experience how much more to over expose form the indicated histogram.

The in-camera histogram won't approximate the raw histogram unless you do some calibration. Start here:

http://blog.kasson.com/?page_id=2459

I'm happy to answer questions, and there are others on this forum who are experts in this.

Jim

[Vladimirovich]

Has anyone figured out a good manual exposure protocol. 

http://www.rawdigger.com/howtouse/calibrate-exposure-meter-to-improve-dynamic-range

[digitaldog]

Just setup a controlled scene where you have a good spectrually neutral white balance tool, add other objects, especially light objects who's detial you do wish to retain. Bracket ⅓ or ½ stop over what the incident meter recommends till you're a good 2 stops over. Bring the raws into the raw converter of your choice as you must understand both exposure and development, the development happening within that converter. Examine what it takes to normalize the rendering. Go too far, you can't bring back highlight detail you desire, you over exposed. ETTR is about getting the most exposure without blowing out highlight data you wish to retain. You'll get a good idea of how far you can go doing such a test.

[Vladimirovich]

ETTR is about getting the most exposure without blowing out highlight data you wish to retain.

with Sony's lossy raw compression ETTR shall be quite moderate... Sony is greatly lossy compressing info in highlights... so while you are gaining in S/N overall you are loosing in gradations by moving certain details close to saturation... this may or may not be important for OP, but he shall remember that.

[digitaldog]

with Sony's lossy raw compression ETTR shall be quite moderate...

OK. But I'd still shoot a test to understand better, within my raw converter, what to expect. Having a +1/3rd stop adjustment over the assumed behavior would still be 1/3rd stop closer to ideal.

When I shot transparency film, I sure wouldn’t want to be off 1/3rd stop.

[robdickinson]

I still dont understand why only leica offer histogram from raw data.

I use magic lantern on canon and it works really well for some of this, but nothing similar for sony a7r.

[Telecaster]

When I shot transparency film, I sure wouldn’t want to be off 1/3rd stop.

With transparency film you had narrow dynamic range (though very fine tonal gradation within that range) and little ability to recover non-optimal exposures. With current sensors it's a different world. Using early digital cameras I was very careful about exposure. Strict ETTR, etc. Nowadays...not so much. IMO as long as you're not careless there's no need to be so fussy and no real-world benefit obtained by being so.

Anecdote: while traveling in the Middle East in the mid-1980s I met an Israeli photographer who exposed Kodachrome 64 at EI 80, then duped his "selects" onto medium format neg film (forget which one) with the highlights pulled up just so. He then printed from the dupes. The prints retained much of the K64 look, not as punchy as Cibachome but with wider DR. A different technique from a different time, though.

-Dave-

[Guillermo Luijk]

I still dont understand why only leica offer histogram from raw data.

Because most users don't care about RAW histograms, or even don't know about their existence.

[ErikKaffehr]

Hi,

Please describe how. I checked the algorithm an I don't think it affects highlights, but if you have done a careful analysis, please share it with us.

Best regards
Erik

with Sony's lossy raw compression ETTR shall be quite moderate... Sony is greatly lossy compressing info in highlights... so while you are gaining in S/N overall you are loosing in gradations by moving certain details close to saturation... this may or may not be important for OP, but he shall remember that.

[JimVehe]

A creative style of Deep and a contrast setting of -3 provides a more useful live view histogram.  When I used those settings the live view histogram was about a 1/4 to 1/3 stop under what the RAW file has when opened in ACR or Lightroom.  The blinkies on playback are still useless. 

[JimVehe]

Thanks Fine_Art.

[Vladimirovich]

Please describe how. I checked the algorithm an I don't think it affects highlights, but if you have done a careful analysis, please share it with us.

not me, but the person who writes the code for rawdigger spent quite some time w/ the details = http://blog.lexa.ru (not in English) ... nothing close to the end of the world, but highlights affected still... a thing to remember.

[alextutubalin]

Please describe how. I checked the algorithm an I don't think it affects highlights, but if you have done a careful analysis, please share it with us.

(I've come here because I see traffic to my blog. I think, it is better to discuss in english than try to translate my russian notes using automated translators

Due to tone curve (highlights compression), the tone distribution between photographic stops is very specific on Sony cameras (against 'common' 12- or 14-bit linear ones, such as Canon or lossless Nikon files).

The topmost stop (from 1EV below camera maximum to maximum) contains only 237 levels.
Next one: 295, next one: 335, and 4th one: 377.
Then tone curve becomes linear, so beginning from 5th stop to shadows the distribution is 'usual' two times decline on each stop (266, 133, 66 33....)

So, for low contrast scenes you should expose your subject 3-4EV below saturation to get more levels (values) on it.
High contrast scenes is another story: you may want to open shadows by turning exposure 'to the right'

You may extract Sony tone curve using RawDigger in CSV format (Menu - File - Dump RAW Curve) and examine it.
Please note:
 1) you should subtract black level (bias). It is 512 for most Sony cameras
 2) Tone curve maximum is 17220 (16708 after black subtraction), while real maximum seen in data is 16116 (after black subtraction), so real tones starts from level #4058)
 3) Tone curve is  12 bit input => ~14 bit output (real maximum is slightly above 2^14 bit), but only even lines are really used: the  raw data (before curve) is 11-bit, so the values are multiplied to 2 before appyling the curve.
 
 4) I've examined tone curves from several Sony cameras (A7R, A99, A900 in cRAW mode, NEX-C3, NEX-7) and these curves are the same. It is possible (but unlikely), that other Sony camera models uses (slightly?) different tone curves.

[ErikKaffehr]

Hi,

Thanks for the explanation but I don't buy it, here is why:

I don't think the reason to use ETTR is to maximise the number of levels in the highlights as there are plenty of them anyway but to maximize exposure in order to:

- Minimize shot noise
- Move the noise floor down

If we assume an uncompressed signal noise will be 126 at 16000 (standard deviation will equate sqrt(number)). So shot noise will be spread over a significant number of levels.

topmost stop

8000 numbers - 237 levels bin width 33.7  noise 126
4000 numbers - 295 levels bin width 13.5  noise 89.4
2000 numbers - 335 levels bin width 5.97  noise 63
1000 numbers - 376 levels bin width 2.65  noise 44
500  numbers - 266 levels bin width  1.9   noise 31

I am pretty sure noise will dominate over banding at any exposure. I don't know why they are doing this, but I guess that "Bionz" works on 12 bit data and compression allows Sony put wider data trough a 12/11 bit processing path.

Noise to signal ratio is proportional to sqrt(signal), so 4 stops underexposure decreases SNR by a factor of two.

I just say that I don't see the Sony compression is a good thing, rather I say I don't think it is the evil it seems purported to be.


Best regards
Erik

(I've come here because I see traffic to my blog. I think, it is better to discuss in english than try to translate my russian notes using automated translators

Due to tone curve (highlights compression), the tone distribution between photographic stops is very specific on Sony cameras (against 'common' 12- or 14-bit linear ones, such as Canon or lossless Nikon files).

The topmost stop (from 1EV below camera maximum to maximum) contains only 237 levels.
Next one: 295, next one: 335, and 4th one: 377.
Then tone curve becomes linear, so beginning from 5th stop to shadows the distribution is 'usual' two times decline on each stop (266, 133, 66 33....)

So, for low contrast scenes you should expose your subject 3-4EV below saturation to get more levels (values) on it.
High contrast scenes is another story: you may want to open shadows by turning exposure 'to the right'

You may extract Sony tone curve using RawDigger in CSV format (Menu - File - Dump RAW Curve) and examine it.
Please note:
 1) you should subtract black level (bias). It is 512 for most Sony cameras
 2) Tone curve maximum is 17220 (16708 after black subtraction), while real maximum seen in data is 16116 (after black subtraction), so real tones starts from level #4058)
 3) Tone curve is  12 bit input => ~14 bit output (real maximum is slightly above 2^14 bit), but only even lines are really used: the  raw data (before curve) is 11-bit, so the values are multiplied to 2 before appyling the curve.
 
 4) I've examined tone curves from several Sony cameras (A7R, A99, A900 in cRAW mode, NEX-C3, NEX-7) and these curves are the same. It is possible (but unlikely), that other Sony camera models uses (slightly?) different tone curves.

[Vladimirovich]

I don't think the reason to use ETTR is to maximise the number of levels in the highlights

and nobody was claiming that ETTR is about the number of levels (since that first article on LL), the claim was that there might be some negative results in some cases w/ some strong postprocessing with Sony's compression scheme if you always expose mechanically to the right as much as possible...

[alextutubalin]

- Minimize shot noise
- Move the noise floor down
If we assume an uncompressed signal noise will be 126 at 16000 (standard deviation will equate sqrt(number)). So shot noise will be spread over a significant number of levels.

For scientific correcntess:  sigma (standard deviation) is equal to sqrt(number) only for unity gain (1e/1ADU) mode. For A7R unity gain is somewhere between ISO200 and 400 (have not measured it yet), for lower ISO signal/noise ratio is lower than sqrt(signal).

Also, I agree: if you 'shot right' you'll get more levels and less noise for shadows. Yes. I agree!

BTW, you need to take these things into account:

1) If you use some kind of non-linear color profiles (e.g. Adobe 'twisted' DCP profiles), your colors will be non-linear corrected and you'll need to 'untwist' them.

2) If someone creates usual (linear, e.g. matrix) color profiles using built-in lightmeter, the profile target will be underexposed heavily (relative to your 'ETTR' shot). So, the profile will compensate for shadow non-linearity (due to noise).
If you apply this profile to ETTR shot, the profile will over-compensate (non-existing in ETTR) non-linearities and colors become wrong too.

So, if you want to move from built-in lightmeter 'to the right', you may need to re-profile the camera for your preferred light measuring mode.

[ErikKaffehr]

Hi,

Thanks for correcting me on the math part, I knew I was oversimplifying.

- To the other issues. I presume that any twisting of profiles is done after decompression of data and that the data is contanimated by noise, is that correct?

- Personally, I don't use a light meter just the histogram, aiming to avoid clipping on non specular highlights (*).

The way I build profiles is to make an ETTR shot of a CC-card, check for clipping RawDigger and use DNG Profile Editor. It seems to work for me. Anything wrong with that approach?

Best regards
Erik

(*) To be more exact, I don't use light meter on the P45+. Shoot an image and check histogram and adjust for ETTR. On the Sony Alpha 99 I use light meter but than go for ETTR on the raw histogram. Works decently well for me on both platforms. I regularly check my images with RawDigger and they seem to be just short of clipping.

For scientific correcntess:  sigma (standard deviation) is equal to sqrt(number) only for unity gain (1e/1ADU) mode. For A7R unity gain is somewhere between ISO200 and 400 (have not measured it yet), for lower ISO signal/noise ratio is lower than sqrt(signal).

Also, I agree: if you 'shot right' you'll get more levels and less noise for shadows. Yes. I agree!

BTW, you need to take these things into account:

1) If you use some kind of non-linear color profiles (e.g. Adobe 'twisted' DCP profiles), your colors will be non-linear corrected and you'll need to 'untwist' them.

2) If someone creates usual (linear, e.g. matrix) color profiles using built-in lightmeter, the profile target will be underexposed heavily (relative to your 'ETTR' shot). So, the profile will compensate for shadow non-linearity (due to noise).
If you apply this profile to ETTR shot, the profile will over-compensate (non-existing in ETTR) non-linearities and colors become wrong too.

So, if you want to move from built-in lightmeter 'to the right', you may need to re-profile the camera for your preferred light measuring mode.