Dual ISO - massive dynamic range improvement (dual_iso.mo)

Started by a1ex, July 16, 2013, 06:33:50 PM

Previous topic - Next topic

0 Members and 3 Guests are viewing this topic.

Audionut

Quote from: Audionut on September 11, 2013, 12:44:16 AM
I guess it could be coded so that SNR limits off mean restore previous functionality.  But to be honest, I think that will cause to much confusion for users in the future.

Or maybe a setting in dual ISO, "link to AETTR".  Else dual ISO is always enabled at the settings specified.

RenatoPhoto

Quote from: Audionut on September 11, 2013, 04:42:01 AM
Nothing.  Dual ISO is controlled automatically.  Just turn it on.

Control the midtones/shadows with the SNR limits.  With dual ISO on, it won't blow your highlight detail like previously, it will adjust dual ISO settings as needed to boost the midtones/shadows.

Nothing is not a good answer because after the first picture the camera automatically sets the Base ISO = Recovery ISO and Dual ISO stops working.   So again, there appears to be a problem with Dual ISO module that should be resolved.

The problem is explained here:

Quote from: RenatoPhoto on September 10, 2013, 04:10:09 AM
Camera tested 5D3
I tried the latest commit (da752ca33b18) and the when using the AETTR + Dual ISO I get a problem. 
I am setting the Dual Iso to +3EV
The first picture I take shows as dual ISO possilbly at base iso + 3EV
The next picture the Dual ISO module has the base iso = recovery iso.  So no Dual Iso Picture is take.
Repeatable.

And here....

Quote from: RenatoPhoto on September 10, 2013, 11:22:12 PM
When I take a photo with AETTR 02% h.i. and SNR OFF I get
DR = 10.9
Same scene with Dual Iso only with Recovery ISO = +3 EV I get
DR = 13.4

Now I turn on AETTR 0.2% h.i plus and SNR = OFF + Dual ISO with Recovery ISO = +3 EV I get
DR=10.9  and Base ISO = Recovery ISO  DUAL ISO is turned OFF automatically

Now I set DUAL ISO TO 800 ISO plus the AETTR as above and I get
DR=10.9 and Base Iso = Recovery ISO , my ISO 800 has been automatically replaced by base ISO

Conclusion:  Dual ISO + AETTR is not working properly.  If I roll back to Sept 07 build then the results are totally different and DUAL ISO works as expected!

and here:

Quote from: RenatoPhoto on September 11, 2013, 04:12:11 AM
I want to keep all of the highlights, that is why I turned off SNR.  So my only option to conserve the highlights is Midtones SNR = 1 if I want to use Dual Iso with Recoveru ISO +3EV
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

Audionut

Quote from: RenatoPhoto on September 11, 2013, 02:38:36 PM
Nothing is not a good answer because after the first picture the camera automatically sets the Base ISO = Recovery ISO and Dual ISO stops working.   So again, there appears to be a problem with Dual ISO module that should be resolved.

Can you post 2 cr2s please.  The first shot where you had your initial settings, and the second shot where dual ISO was reset to recovery ISO = base ISO.

RenatoPhoto

Quote from: Audionut on September 11, 2013, 03:35:20 PM
Can you post 2 cr2s please.  The first shot where you had your initial settings, and the second shot where dual ISO was reset to recovery ISO = base ISO.
Would love to do it but I dont have the bandwidth to do that.  What should I look for in the cr2.  maybe I can analyze it.  I know what dual iso is and there is no dual ISO on the second image because DUAL ISO says so, "Both ISO are identical, nothing to do"
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

a1ex

If the first CR2 has a SNR higher or equal to what you have requested in menu, ETTR will disable dual ISO. This is normal and not a bug.

RenatoPhoto

Quote from: a1ex on September 11, 2013, 03:45:32 PM
If the first CR2 has a SNR higher or equal to what you have requested in menu, ETTR will disable dual ISO. This is normal and not a bug.
SNR=OFF
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

Audionut

What highlights are or are not being blown in each of the photos.

Does setting SNR limits correlate with the description provided by a1ex 2 pages back. 
Are the highlights being blown with SNR limits enabled.
Set SNR limits to 8/3EV midtones/shadows.  Are the highlights being blown.


Always take enough photos after changing AETTR settings to ensure convergence has been met.

a1ex

Had a chance to play with it today. I've set it on Auto ETTR (SET button in LiveView), highlight ignore 0.1%, midtone SNR at 7 EV and shadow SNR disabled.

In most cases it worked quite well. Only had some minor problems:

1) when ISO reached 6400 (it was undecided between 400/6400 and plain 6400 - quite a big diference). The problem appears only when it's too dark and exposure limits are reached (extra DR needed is estimated incorrectly).

2) in "always on" mode, confirmation beep was not always right (overall exposure didn't change, but dual ISO range did; however, the "beep system" didn't notice that). So, after getting a single beep, I had to take one more picture.

LiveView + SET worked fine.

3) sometimes it chose 3200/6400, and there was probably no benefit in doing that. Of course, I could have limited it to 1600 or 3200, but I also wanted to watch the algorithm making the choices.

Besides these small quirks, I think ETTR is now doing near-optimal exposure for dual ISO.

I may post some pics if there'll be anything worth showing.

Audionut

There appears to be a problem with the noise calculation for ISO 6400 (for the 5D3 at least).
The reported noise level continues to drop for all dual ISOs until ISO 6400 recovery.

QuoteInput file     : _UAL6623.cr2
Canon EOS 5D Mark III detected
White level    : 12500
Black level    : 1963
Noise levels   : 10.65 6.10 6.26 10.49 (14-bit)
ISO difference : 3.97 EV (1571)
Black delta    : 2.39
Dynamic range  : 10.75 (+) 9.95 => 13.92 EV (in theory)
Hot pixels     : 64
Cold pixels    : 378
Noise level    : 2.17 (16-bit)
Dynamic range  : 14.25 EV (cooked)
Black adjust   : 1
Output file    : _UAL6623.DNG

QuoteInput file     : _UAL6624.cr2
Canon EOS 5D Mark III detected
White level    : 12500
Black level    : 1961
Noise levels   : 16.08 6.13 6.22 15.70 (14-bit)
ISO difference : 4.91 EV (3006)
Black delta    : 1.01
Dynamic range  : 10.75 (+) 9.36 => 14.27 EV (in theory)
Hot pixels     : 5
Cold pixels    : 419
Noise level    : 1.66 (16-bit)
Dynamic range  : 14.63 EV (cooked)
Black adjust   : 1
Output file    : _UAL6624.DNG

QuoteInput file     : _UAL6625.cr2
Canon EOS 5D Mark III detected
White level    : 12500
Black level    : 1954
Noise levels   : 26.39 6.21 6.34 26.81 (14-bit)
ISO difference : 5.87 EV (5828)
Black delta    : -0.31
Dynamic range  : 10.73 (+) 8.62 => 14.48 EV (in theory)
Hot pixels     : 4
Cold pixels    : 649
Noise level    : 3.22 (16-bit)
Dynamic range  : 13.68 EV (cooked)
Black adjust   : 1
Output file    : _UAL6625.DNG

This occurred in 3 different scenes with varying levels of DR.    DxO results suggest that the advantage of ISO 6400 over ISO 3200 should be neither here nor there, and my own testing showed a very minimal gain (Std Dev continued to drop for a selection of flat textures in the scenes).

https://www.dropbox.com/sh/di2lkp2knv5bqso/XPN9ptKf-Y

edit:  The results are for base ISO 100
@ base ISO 200, the noise level only increases slightly for recovery ISO 6400.
@ base ISO 400 and 800, the noise level continued to drop for recovery ISO 6400 and increase for recovery ISO 12800.

N/A

Getting some damn great pictures on the 600d, amazing work gentlemen.
7D. 600D. Rokinon 35 cine. Sigma 30 1.4
Audio and video recording/production, Random Photography
Want to help with the latest development but don't know how to compile?

Khairilasny

Quote from: N/A on September 12, 2013, 02:59:05 AM
Getting some damn great pictures on the 600d, amazing work gentlemen.

2nd that!!

optik

Is it me or the app please help I'm using  cr2hdr // Beta 1.0 app to convert dual ISO I have success  in windows but preferred to use Mac but this is the error that I keep getting

cr2hdr // Beta 1.0


Input file : 6I0A5237.CR2
sh: dcraw: command not found sh: dcraw: command not found Error: dcraw output is not a valid PGM file Logfile    : 6I0A5237.txt

THE END

Keep up the incredible work that you guys do and if you're working too hard then definitely get rest...
Canon 5D Mark III, Canon 24-105mm F4, Canon 50mm F1.4, Nikkor 35mm f/1.4 AI-s, Tokina AT-X 16-28mm F2.8 Pro FX, 2 Lexar 1000x 64GB, Kumputer Bay 1000x 64GB, FCPX, Nuke, Modo, AE,
DaVinci

fpena06


RenatoPhoto

I have done some test to try to understand how the new link between AETTR and DUAL ISO is behaving.

As I stated before, the new system when using the SNR = OFF or low SNR numbers, always defaults to Base ISO = Recovery ISO.  My explanation to this behavior is that the modules is doing all the calculations and deciding automatically that DUAL ISO is not needed and therefore making Base ISO = Recovery ISO.

So now I will test this theory as follows:
1. Set AETTR to 0.2% h.i. and SNR = OFF and took some pictures until the exposure converges.
2. Turn AETTR OFF (to maintain its corrected exposure) and use DUAL ISO to manually test different iso combinations.
3. Compare these results with AETTR+DUAL ISO at different levels of SNR

First is the manual method:

In these images it is obvious that the shadow noise is improved with the use of DUAL ISO

Now lets look at the new AETTR + DUAL ISO

We see that the system only responds to some very high SNR numbers 8/3 to reduce the noise in the shadows, but with lower numbers that shadows are totally ignored and DUAL ISO is disabled automatically. 
Also note that the AETTR with SNR=8/3 activated the DUAL ISO at 100/400 and this results where similar  to the simple 100/400 DUAL ISO test done first.   At  maximum SNR = 8/4 I was able to get 100/1600 dual ISO but it took a few tries until I got most of the DUAL ISO options to show up with the new system.

Conclusion is that the new system will use the DUAL ISO technology when needed to maintain the h.i. setting (0.2% in this test) and reduce the SNR shadow and midtones.  A similar result can be obtained by first doing the AETTR, disabling AETTR, and enabling DUAL ISO with manual settings.  The main difference is that the new system will search automatically for the DUAL ISO needed but the SNR numbers have to be very high for these parameters to make effect in the DUAL ISO.  I had used AETTR and never used 8/4 to obtained good results, I typically used 4/2 but at these levels DUAL ISO is not enabled.  This was my main conflict in thinking that it was not working but I just had to raise the SNR to the limits to get it to work.

This analysis then led me to explore what SNR settings will allow me to exploit the DUAL ISO benefits.  I found the following combinations that will give me what I am looking for in DUAL ISO performance with full sun and shadow outside setting.

SNR = 8/4 = Dual ISO = 100/1600
SNR = 7/4 = Dual ISO = 100/800
SNR = 6/3 = Dual ISO = 100/400
SNR = 5/3 = Dual ISO = 100/200

Now  I switched to a full sun and lots of full shadow through a window ( inside my house):

SNR = 8/4 = Dual ISO = 100/3200
SNR = 6/2 = Dual ISO = 100/1600
SNR = 3/1 = Dual ISO = 100/800
SNR = 2/1 = Dual ISO = 100/400
SNR = 1/1 = Dual ISO = 100/200

Now this leads me believe that we basically have maximum six levels of use for DUAL ISO when the condition are adequate.  If we have lower light situations where the base ISO is lower we lose one lever for each base ISO increase.  For example if we have base ISO of  200 we have five levels l (200/400, 200/800, 200/1600, 200/3200, 200/6400), for base iso = 400 we have four levels of adjustment (400/800, 400/1600, 400/3200, 400/6400).

This makes me think that we could simplify this system and present the user with just the number of shadow recovery options based on the base iso instead of having all of these SNR options.  For example if base iso is 100 then the allows adjustments are from 1 to 6.  If the base ISO=200 then we have only 1 to 5, etc.

Right now I am presented with 32 possible adjustments ie 8/4, 8/3, .... 7/4, 7/3,  6/4, 6/3 and some of them will not produce any apparent results.  Now if this could be reduced from 32 to 6 options then ease the usability will be much greater.

Just a final note the highlights have been independently maintained without overexposure which lead me to believe that are independently controlled by highlight ignore.  Will probably need a separate test.

And finally I thank you Alex for this amazing technology and I do appreciate all the hard work that you do.

P.S.  I am learning  from doing these test so please correct me if I did something wrong!
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

Danne


a1ex

@RenatoPhoto: that's way too complicated. Simply state:

a) how much noise you are OK with (let's say 6EV or 7EV)

b) how many highlights can be ignored (0.1% is a good overall choice, 0.5 for something like a fairly large window in a dark cave where not even 20 stops of DR are enough, and 0% if you don't have any specular highlights).

Then let ETTR do the rest. That's what I do.

Audionut

I've been exploring the idea of using dual ISO for quasi ISOless mode.  Base ISO is always ISO 100, with adjustments to recovery ISO only.  ETTR+dual ISO is fine for still shots, but for running and gunning under varying light conditions, it is much to slow.

In a quasi ISOless mode, I could set dual ISO 100/400 and cover an exposure value range of 6-16EV in Av mode (1/125-F/1.4 with this lens).  If I didn't care about my highlights so much, I could use dual ISO 100/1600 and cover an exposure value range of 4-16EV and deal with the reduced resolution in post.  With dual ISO 100/400, this also allows 2 EV of underexposure leeway for the base ISO.  ie: not only do I not need to worry about ISO adjustments (as much), I don't have to worry about being extremely accurate with my exposure to ensure my base ISO is always ETTR.  The recovery ISO takes up the slack ;)


Some caveats with the current functionality.
The histogram always uses the lowest ISO.  Where recovery ISO becomes the dominant brighter exposure, the histogram becomes unreliable.  ETTR hint does not work, and the histogram will no longer display overexposure results.
It doesn't appear as if base ISO is boosted in anyway (in cr2hdr) to fill in the exposure gaps in the recovery ISO, where recovery ISO is brighter.  Suggestion, where recovery ISO is brighter then base ISO, determine white level of base ISO, boost the exposure of the base ISO and use that data for AA correction in the recovery ISO. It's better to have some accurate data, even if that data has a higher noise level, then to interpolate missing data.
This could also be handy with the current functionality, to fill in any resolution loss where base ISO is not ETTR.

a1ex

I don't get it; cr2hdr only cares about higher iso and lower iso (doesn't matter which is base and which is recovery - this one is mostly relevant when using LiveView). So 100/1600 and 1600/100 are processed identically.

Of course, these two might have slightly different noise characteristics because Canon sets some more parameters, not just amplifier gains (maybe NR or who knows what else). In practice I didn't notice much difference.

The histogram always uses the lowest ISO for ETTR reasons (so you can see how far you can push it to the right). If something is covered by at least the lowest ISO, it's not overexposed.

About ISO-less mode: I found that for disco lights (changing all the time, but overall pretty dark with some very bright areas), 1/30 (slower than that results in shake) and ISO 100/1600 covers pretty much everything (some sort of auto that actually gets it right).

Audionut

The test shot I was looking at had some overexposure that I did not determine during capture.  I then assumed that the resolution loss in the highlights was due to not using the darker exposure  to fill in the resolution loss of the brighter exposure.  I guess because I am always thinking of base ISO as the darker exposure.

RenatoPhoto

Quote from: a1ex on September 12, 2013, 07:12:59 PM
@RenatoPhoto: that's way too complicated. Simply state:

a) how much noise you are OK with (let's say 6EV or 7EV)

b) how many highlights can be ignored (0.1% is a good overall choice, 0.5 for something like a fairly large window in a dark cave where not even 20 stops of DR are enough, and 0% if you don't have any specular highlights).

Then let ETTR do the rest. That's what I do.

Thanks for your reply,

I did the first photos in manual mode to explore the area that the new AETTR + DUAL ISO does not allow me to explore. 
For what purpose?  So I can compare these iso combinations and results with the results of AETTR + DUAL ISO. 
Why do I want to explore this area?  Because three days ago I could do it by simply putting my Recovery ISO = +x EV.  Now I cant get these results because the new module just decided that this is not what should be done and sets BASE ISO = RECOVERY ISO based on the SNR settings.

Request: Can the DUAL ISO module be configured with an option to do the automatic ISO calculation and one option to do what it did before?  i.e.  Auto Dual Iso (ON - OFF)?  This would give this module the flexibility to take your optimization or not.
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

a1ex

Show me a scene that requires dual ISO, but auto ETTR said it doesn't.

RenatoPhoto

Quote from: a1ex on September 13, 2013, 12:36:31 AM
Show me a scene that requires dual ISO, but auto ETTR said it doesn't.
In the first row of images above, all of them some of them have lower noise than that obtained by AUTO ETTR + DUAL ISO.
Why?  Because I am using higher ISO combinations that the new module does not allow bases on the  calculations.
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

RenatoPhoto

Let me correct that, I get lower noise with higher ISOs that the new module does not allow for the scene.  For example I the 100-800 dual iso image is lower in noise than the 100-400 which was the maximum iso range that the module allowed for this scene.
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X

Audionut

In the images you have AETTR SNR 8/3 @ ISO 400 for auto recovery.  Why did you not have SNR 8/4?
You then go on to show in other scenes (with no images), where setting SNR 8/4 resulted in much higher recovery ISOs.

RenatoPhoto

Quote from: Audionut on September 13, 2013, 03:06:33 AM
In the images you have AETTR SNR 8/3 @ ISO 400 for auto recovery.  Why did you not have SNR 8/4?
You then go on to show in other scenes (with no images), where setting SNR 8/4 resulted in much higher recovery ISOs.

Yes in this scene I din not think it was required to max out the SNR but later I realized that you could get higher ISO if you max out at 8/4.  Under similar conditions as the presented photos I was only able to get 100-800 ISO but no higher as I wished to test.

In similar tests the obtained result under similar ISO are similar but the point I am trying to make and seems difficult to understand is that a user may want to test higher ISOs which are now not available under automated scenarios of AETTR + DUAL ISO.  These scenarios where quite easy to use by allowing the ISO module to set the Recovery ISO as a relative number of +xEV.  I personally think this is good to leave in the module.  Currently the +xEV option is not working in Aettr.
http://www.pululahuahostal.com  |  EF 300 f/4, EF 100-400 L, EF 180 L, EF-S 10-22, Samyang 14mm, Sigma 28mm EX DG, Sigma 8mm 1:3.5 EX DG, EF 50mm 1:1.8 II, EF 1.4X II, Kenko C-AF 2X