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

[Danne]

I tried commandline option for dng converter "-l Output linear DNG files."


http://wwwimages.adobe.com/content/dam/Adobe/en/products/photoshop/pdfs/dng_commandline.pdf

When adobe dng conversion is done through cr2hdr20 it seems to use the default settings for compressed files. Works well and one don,t notice differences in CR2 files but for movie sequences one gets the same resulting flicker as without -same level option in cr2hdr20bit.
I tried to set same white and black level for the adobe transcoded file but with no success. However with setting "-l"  I get total linear results.

[N/A]

For the 70D, cr2hdr needs to be compiled from nikfreaks repo.

https://bitbucket.org/nikfreak/magic-lantern/src/ac3023b990592a47c15ff3c3cdd3b85b3f320f96/modules/dual_iso/?at=70D

Here's a compile from the 70D branch along with a dcraw compile.

https://www.dropbox.com/s/37sn9wwhkrwedk9/cr2hdr-70d.zip?dl=0

[Danne]

Thanks!

[rltksun]

Hi... I USING 70D AND ML INSTALLED.
I have problem now.
I don't know how to use unix file for 70d dual iso...
(https://www.dropbox.com/s/37sn9wwhkrwedk9/cr2hdr-70d.zip?dl=0) <- this
NEED HELP!!!

[N/A]

Are you on a Mac? Open a terminal window, drag and drop cr2hdr into it, then drag your dual iso cr2 into the window and press enter.

[arhi]

just tried dual_iso on my 50D (build from yesterday - 5th April) and I have some issues post processing it ...

often I get this error (Interlacing method not supported, ISO blending didn't work):

Code Select Expand


e:\bin\cr2hdr>cr2hdr.exe  in\IMG_5704.CR2
cr2hdr: a post processing tool for Dual ISO images

Last update: d0ac769 on 2014-01-23 10:13:39 UTC by alex:
cr2hdr: experimental white level routine with two separate levels ...

Active options:
--amaze-edge    : use a temporary demosaic step (AMaZE) followed by edge-directed interpolation (default)
--cs2x2         : apply 2x2 chroma smoothing in noisy and aliased areas (default)

Input file      : in\IMG_5704.CR2
Camera          : Canon EOS 50D
Full size       : 4832 x 3228
Active area     : 4770 x 3177
Black borders   : 62 left, 51 top
Black level     : 993
ISO pattern     : BdBd GBRG
Interlacing method not supported
ISO blending didn't work


other time the cr2hdr works ok but the resulting DNG file is identical to CR2 file (the interlacing dark/light lines)

all three files (IMG_5704.CR2 that won't go trough cr2hdr, IMG_5703.CR2 that converts into bad IMG_5703.DNG) available here http://elco.crsndoo.com/files/ML/ML_dual_iso_problem_50D.zip  (62M archive)

EDIT: the "Experimental 20-bit version: cr2hdr-20bit.exe" for e.g. reads this IMG_5704.CR2 properly and the output DNG is looking good

[a1ex]

@arhi: are there any files that fail in the 20-bit version?

[arhi]

@a1ex - not so far, 20bit version so far worked perfectly

[Myxir]

Hi!

At first: Thanks for doing all this work for Canon (Magic Lantern) community!

I'm  using a 1100D with ML. I tried DualISO over the last few days to see how it works. But I'm absolutely unable to reproduce results such as shown in the start post. I took two  shots, one with DualISO (100/1600 or so) an one just with 100. Then  I copied to my PC and converted Dual CR2s with cr2hdr + barracudaGUI to DNG. But in RawTherapee there's almost no visible difference between DualISO.dng and NoDualISO.cr2!?
Even after Tone mapping to the extreme, all I get is a little bit less noise in the shadows. I looks like 0,3 EV (or so) more dynamic range, not like 2,5... :/
Before convertinmg to dng, the dualiso.cr2 looks better, with good exposed shadows.

Did I make something wrong or forget something? Please help...

[Danne]

Post the originals.

[Myxir]

~40 mb? Dropbox?

[Danne]

yes

[Myxir]

-> Dual ISO Test

No great picture, I just wanted to have much contrast...

[Audionut]

The dual ISO image has the edge in noise.  Although the noise in the CR2 increases the perceived sharpness for me.

The contrast isn't so bad in this image, you can tell easily because the regular CR2 is pretty clean.  If you shot this same exposure with flat textures, the difference would probably be perceived as being greater.  To put dual ISO through it's paces, find something with even greater contrast (maybe underexpose a little if you can't find a scene).

Here is an example of extreme contrast.

[Danne]

I played around with highlight and shadows sliders. Also with tone curve tool in acr. The dual iso file worked fine. Maybe find some more contrasty situations as suggested by Audionut to really see the difference.

[Myxir]

Even in a scene with backlight (Directed towards the sun or in a dark room and bright window) seems not to be soo contrasty.
But yes, when i do an underexposure at -2 or -3 EV theres an extreme improvement in quality. But in which "real world" situations do I have that?
DualISO is working as expected. I just wanted to know that. Thanks...

@Danne/Audionut: I used exactly the same post processing settings for both pictures (Dual iso and single iso). I was expecting results like that: http://www.magiclantern.fm/forum/index.php?topic=7402.0
These scenes doesn't contain too much contrast, too. But now I guess these differences come from different post processing. Otherwise i can't explain...

[Audionut]

But yes, when i do an underexposure at -2 or -3 EV theres an extreme improvement in quality. But in which "real world" situations do I have that?

The suggestion to underexpose was only to simulate a scene with greater dynamic range.  By underexposing you are pushing more of the midtones down towards the shadows, where all of the noise is.  In a large dynamic range scene this is what happens when you try and capture the highlights.

Dual ISO doesn't clean the midtones or highlights, only the deep shadows.  If you don't need to capture deep shadows, stick to regular Canon images as you won't suffer any of the dual ISO drawbacks.

[Myxir]

]The output files contain 16-bit raw data in DNG format, that looks identical to a picture shot
in the same conditions at ISO 100. However, this DNG has a lot less noise in the shadows, and
therefore you can push the exposure a lot higher without getting massive noise (+6EV should
be quite clean).

Oh lol. Maybe I should have read that before. Now it's clear!
Although I think, +6EV is a little bit too much for my 1100D, even with DualISO...

By the way: If I change the DualISO settings from 100/800 to 800/100 it should save the highlights instead of shadows, right?

[Audionut]

Yes, expose for the shadows and dual ISO will recover highlights.

[Myxir]

Okay, perfect.

A few other questions:
Why can't I use ISO 6400 for Dual ISO? Even if I select 6400 from Canon menu and Dual ISO menu it says "3200/3200 -> Both isos are the same".
And what is if I select Highlight tone priority? ML says Canon analog ISO is 100, equivalent ISo is 200, but DuaISO says 100/x, not 200/x. What does that mean?

[Licaon_Kter]

I can confirm that you can't set up 3200/6400 ( it will reset to 3200/3200 thus disabling it ), but you can set 6400 in ML and it yields 6400/3200.

[Audionut]

HTP is only useful for JPG/H.264.  With raw it simply underexposes one full stop.

ISOs above 3200 with dual ISO are pointless.  For further information, start at the first post of this thread and enjoy the read.

[Myxir]

Because when you increase ISO from 3200 to 6400 you "recover" 1EV of shadows but lose it on overall dynamic range (so you get nothing)? Even from 1600 to 3200 there's only 0.1 EV DR more...

Are there differences between different camera models? Because full frame cameras for  example are a lot better at high ISOs...

[Audionut]

http://www.magiclantern.fm/forum/index.php?topic=10111.0
In the first post of that thread, also check the section entitled "Any recommended reading?".  The DxOMark site will list the dynamic range for all tested cameras.

ISO 1600 > ISO 3200 may net you 0.4 EV on some camera when you are using standard image capturing.  With dual ISO you shouldn't expect to see that result.  Each ISO is only half resolution, so you throw away one full stop of (full) midtone resolution, to "gain" 0.4 EV of half resolution shadow detail.
When you use recommended settings (100/800....100/1600), since the top couple of stops are generally white anyway, the resolution loss may not even be noticed.  But when you start using extreme settings (100/3200), you're throwing away another full stop of midtone redolution, and this resolution loss is smack bang in the midtones.  You probably want all available resolution for the midtones.

Where you use a low base ISO of say 100, the situation is probably even more dire.  Here, you probably have your exposure (shutter/aperture) set for the base ISO.  Since ISO doesn't effect shot noise, not only are those last few points of EV only half resolution, but they are also heavy with shot noise.

[winfel]

There seems to be an issue with the created DNGs in Lightroom 6

https://www.dropbox.com/s/sj7htx3tr57y4op/Screenshot%202015-04-24%2017.12.37.png?dl=0

Shows the same created DNG file in Lightroom CC and in Lightroom 5. As you see, the interpretation is completely different. In the CC case everything is overexposed and cannot be brought back. I created the DNG an a Mac out of a Dual ISO file made by a 70D.

The original DNG file can be found in here: https://www.dropbox.com/sh/dnoiuhnqanjcbha/AADccyx0pHU2IiZhBCofNkM2a?dl=0

Does this occur with other Dual ISO DNGs as well?