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 2 Guests are viewing this topic.

dmilligan


ursamajor

In the first post of this thread a1ex said:

Histograms, ETTR, playback tricks

- ETTR integrates with dual ISO (just enable both). It will use dual ISO only in high dynamic range scenes. Adjust the SNR limits to let the algorithm know what exactly is a "high dynamic range scene" and to fine-tune the balance between midtone/shadow noise and highlight aliasing.


I enabled both DualISO and ETTR on my 550D, but even in normal scenes I get DualISO RAW. Where am I wrong?

Audionut


ursamajor

Is there a detailed guide somewhere for how to use these features or it's only in ML general manual?

Audionut

The best we have is this:  http://www.magiclantern.fm/forum/index.php?topic=11269.0

Auto ETTR is pretty well documented, just follow the links.

ursamajor

Very good, thank you very much!

Trimis de pe al meu X2_Soul folosind Tapatalk


jpegmasterjesse

Hi guys, what's the latest with the 5d Mark II Dual-Iso stuff?

Is there a newer version of the dual-iso.mo?

Is there a better workflow than MLP?  I wasn't able to find the hacked CR2HDR.

visiono

my canon 50d has the lowest iso at 200. Since multiples of 160 are ideal for less grain does this mean my lowest option in real iso is 320?

Also, canon digital iso says highly recommended 0, however i can never get it below 1. is this due to 200 iso limit by canon ?

So, if i don't want to go any higher than 1600 my best setup is 200/1600 ?
i was able to get real iso as low as 50? are there disadvantages/advantages to this is relation to dual iso?


Walter Schulz

You may want to consult your manual about turning off HTP (Highlight Tone Priority) and revisit ISO settings after.

Audionut

On the 50D, the lowest native ISO appears to be ISO 200.  There's some ISO results for the 50D in this thread.  That thread also details things regarding ISO multiples of 160.

Sooner266

Hey guys,
I've used ML for several years and have been getting more into timelapses recently, using auto ETTR to shoot my first holy grail sequence which turned out beautifully. The past few days I started adding dual ISO to the mix and, while the individual images look incredible, I've been having some issues getting it to work seamlessly in timelapses.

I've spent the past couple days searching the forums but I haven't found my issue addressed, which is: doesn't allowing ML to turn dual ISO on and off as needed introduce an incorrectable color/exposure change in the final result due to the extra processing step (cr2hdr)? I just got back from a test holy grail timelapse and compared sequential (10s) images before and after ML enabled dual ISO as it got darker. With identical exposure and raw settings, the two images show a huge change in color and exposure. Also the dual ISO images have a few more pixels (5202x3465 vs 5184x3456). I know my SNR values were probably wrong as it actually disabled dual ISO as it got even darker, but I think the question is still valid in changing light conditions.

I'm shooting on a T2i using the latest ML (2016Jul09), and BarracudaGUI pointed at the cr2hdr executables from the latest zip. First I tried feeding the raw CR2s into Barracuda, and the dual ISO images were extremely different. Then I read someone on the forums say they converted all CR2s to DNGs with ACR, then Barracuda, which turned out closer, but still noticeably different.

Has no one else encountered these problems? Am I just missing something in shoot settings or in post?

Here's a comparison of two of these images, both converted to DNG, run through cr2hdr, and set to same raw settings:  http://i.imgur.com/Rg0LpE0.jpg

Thank you for these incredible tools, can't wait to get this working!

a1ex

I have shot a few timelapses with mixed dual/regular ISO, but didn't manage to render them yet, so I'm not able to fully answer your question right now.

Your sample looks like it needs some deflicker. Try this script: http://www.magiclantern.fm/forum/index.php?topic=8850

If it doesn't do the trick, you might also need a black level adjustment (which can be found with a linear fit), or (hope it's not the case) you might have to use some sort of histogram matching (if the response curves are not exactly linear).

If all else fails, shooting all the pictures in a timelapse at ISO 100/1600, and letting ETTR adjust only the shutter, should result in a set of images easier to deflicker. The --same-levels option in cr2hdr might also help.

These are just my expectations, as I didn't actually try these settings in real-world scenarios.

Sooner266

Thanks for the info. Yes, it needs deflicker, but I'm quite sure the flicker is due to switching to dual ISO; the images on either side of the transition show absolutely no flicker. It's more than an exposure change though; if I correct the brightness I can get it very close, but there is still a subtle difference in color.  If possible I want to try to find shoot settings that don't cause that flicker to begin with and give me a very linear post process workflow, so the last option sounds like exactly what I'm looking for. How would I force ETTR to only adjust the shutter without changing ISO? Even with link to dual ISO disabled, ETTR still adjusts the base ISO.

jhon_wane

Hello Magic Lantern team,

I'm trying hard to get DUAL ISO to work, without much success.
I must do something wrong in the workflow or i do not understand how to use the feature, although i read a lot of material.

When trying to get most data out of the sensor, I'm shooting with AETTR and dual iso (100-1600).
That way I'm getting a lot of highlight information, but the blacks and shadows are pitch black.
When I try to recover the data, I must pull them all the way to +100 in order to see something in the shadows,
but that of course insert a lot of noise to it. So much i cant use the picture for almost anything.

Am I doing something wrong?
Does the DR is just too big that even DUAL ISO 100-1600 cant handle it?
Should i expouse the image without AETTR?
I'm using 70D,  latest CR2HDR, ACR 9.1.1 or LR 6 (same results).
I want to use it so much, help will be very appreciated.
Thanks.

Here is an example of DNG file (after CR2HDR) I cannot bring to life:
https://www.dropbox.com/s/sdj6zgf8u77xken/DUAL0794_100_1600.DNG?dl=0
Thanks.

a1ex

Quote from: Sooner266 on July 24, 2016, 08:05:14 AM
It's more than an exposure change though; if I correct the brightness I can get it very close, but there is still a subtle difference in color.

I expect a difference in black level - there is some guesswork going on. When switching the ISOs back and forth, the low ISO ends up having a slightly higher black level, and their average is what you get in exif.

If you upload the two raw files that you expect to be identical, I can take a look.

Quote from: Sooner266 on July 24, 2016, 08:05:14 AM
How would I force ETTR to only adjust the shutter without changing ISO? Even with link to dual ISO disabled, ETTR still adjusts the base ISO.

There isn't an option for that right now; the max ISO in ETTR is set in Canon menu, but I doubt you can set it to 100. Editing the source code is probably the best option right now. You can tweak the settings so it will be less likely to select higher ISOs (e.g. disabling SNR and highlight ignore), but this will most likely result in underexposure.

Sooner266

Quote from: a1ex on July 25, 2016, 05:08:50 AM
If you upload the two raw files that you expect to be identical, I can take a look.

Editing the source code is probably the best option right now.

I appreciate it, and I'm sure with enough cr2hdr tweaks we could get those two particular shots to converge, but those tweaks would go out the window if the dual ISO EV ratio (or other settings) changed, wouldn't they? Indeed my lowest max ISO is 400.

Yes I think looking to the shoot code would be a better solution. I'd already forked it to work on a couple more options for ettr timelapses, so I'll also see about locking iso entirely. Just could not get the ubuntu virtual box to compile modules due to its age (not a linux person), so I'm going to try my luck with cygwin.

jhon_wane


a1ex

Quote from: Sooner266 on July 25, 2016, 07:06:32 AM
I appreciate it, and I'm sure with enough cr2hdr tweaks we could get those two particular shots to converge, but those tweaks would go out the window if the dual ISO EV ratio (or other settings) changed, wouldn't they?

Of course; I only wanted to find out what kind of mismatch is there (and whether it can be corrected with only a black level adjustment, or it needs something more complex).

Quote from: Sooner266 on July 25, 2016, 07:06:32 AM
Yes I think looking to the shoot code would be a better solution. I'd already forked it to work on a couple more options for ettr timelapses, so I'll also see about locking iso entirely.

Nice! You may want to look at (and build on top of) this branch: https://bitbucket.org/hudson/magic-lantern/branch/ettr-bulb

DeafEyeJedi

Quote from: jhon_wane on July 26, 2016, 07:47:59 PM
Anyone can help please to my previous post?

Not sure how else to help other than to assume that this was due to an user error? All I think of is that you most likely shot this Dual-ISO under exposed at least in this case since you're aiming for the shadows, right?

Which converter did you use to spit this DNG out? Do you still have the original Dual-ISO CR2?

According to my tests with your file it tells me that you shot this under exposed by 2-3 stops & possible more (I had to push the exposure up in ACR)? I also applied Cinelog-C just to see how much it helps with noise. Fun stuff that you can even use this for Photo work (Thanks @Andy600) and let alone with Dual-ISO even if it wasn't intended for!  ;)



Perhaps next time try to have the ETTR reads indicating 0.0 (rather than -1.3 or +1.50) on your LCD screen while exposing for ETTR. I've never been a fan of AETTR maybe because I'm just plain lazy and refuse to learn how to use it. I mean who wouldn't want to control the lighting manually?

Remember it's better to protect the shadows than the highlights since it's RAW and you can just pull the exposure down in post to preserve the noise levels (even takes away too) while keeping the highlights so I try to keep this method in mind when I go out and about.
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

jhon_wane

Thanks,

I know I must be doing something wrong.
Here is the original file:
https://www.dropbox.com/s/z4y1nw4anvbd5wm/DUAL0794.CR2?dl=0

I use CR2HDR latest versions in order to convert it to DNG (using BarracudaGUI).
Can you please help me and explain what is the best workflow for taking a picture using DUAL ISO? I'm very confused.
I understand that I should expose for the lower (base) ISO.
I thought I need to get all the highlights without clipping (that is why I used AETTR) and the recovery ISO (1600) will take care for the shadows - Am I wrong (apparently - yes..)?
So...
Should I expose for the highlights? shadows?
Should I avoid any clipping - shadows and highlights?

Sorry for the newbie questions,
10x again.

Danne

There,s actually nothing wrong with the file. Try set exposure for correct highlights(dark shadows) and set something like 100-1600 for dual iso. Contrasty situations.
In post pull out the shadows and pull down the highlights. Test, test, test. And post your results.

a1ex

Quote from: jhon_wane on July 26, 2016, 09:22:30 PM
Should I expose for the highlights? shadows?
Should I avoid any clipping - shadows and highlights?

Only you can decide what to clip and what to keep in the picture. Raw zebras will tell you exactly what is completely clipped (pure white), or only one or two color channels clipped (can be recovered if color accuracy is not critical), or too noisy (SNR below some threshold). When you enable Dual ISO, the RAW-based zebras will also tell you which parts of the image are covered only by the lower ISO (that is, where it's likely to get aliasing artifacts).

For ETTR, in this particular case, you could allow it to clip some more highlights (to get better shadow detail). The issue here was the sun being much larger than your highlight ignore threshold.

I have in mind some changes to ETTR that would identify such strong highlights and clip them, but didn't manage to implement them yet. Will keep your CR2 to use it as test data.

jhon_wane

Thanks a lot guys-
for the tips and for your time trying analyze my picture.

I will keep trying and update as soon as I can.
Cheers.

l_d_allan

Id the cr2hdr.exe from Dec. 7, 2013 the "latest / greatest"?

Seems like there was a 20-bit semi-experimental version that was supposed to be compiled. Or is that the Dec 7, 2013 version?

Walter Schulz