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

[jOt]

Well, if you really want to remove the error checks, feel free to do so, but you'll get just garbage.

Hint: I can't troubleshoot anything from these logs (other than reaching the above conclusion).

I attached these logs just to show where it stuck. I have no programming skills, I simply thought that maybe it's a matter of tolerance in ISO difference estimating process. Is it possible to do the conversion in photoshop? Any idea?

[RenatoPhoto]

I attached these logs just to show where it stuck. I have no programming skills, I simply thought that maybe it's a matter of tolerance in ISO difference estimating process. Is it possible to do the conversion in photoshop? Any idea?

maybe you can upload the two cr2 for some one to analyze.

[jOt]

maybe you can upload the two cr2 for some one to analyze.

If somebody is able to convert second file, let me know, please.

http://www.mediafire.com/?kr1che73pz9tmdl
http://www.mediafire.com/?dxoaa17l87dfgec

[NateVolk]

Shot today in DualIso, using stock ETTR settings, pushed Set in LiveView to meter.  Ended up with a bunch of files that can't be converted.  Here's the first one.  I'm using the last experimental update from a1ex a few pages back.  Thanks!

https://docs.google.com/file/d/0B_Id0y3JG-KEQldiSFNTSWpJbWs/edit?usp=sharing

[a1ex]

That's almost completely overexposed in one ISO. Can you reproduce it?

A video of the camera screen when metering the scene should help. None of my ETTR pictures in full sunlight looked like this...

[a1ex]

Latest cr2hdr should handle these shots properly.

[SpcCb]

Last build (2013-09-20, second run) tested on 5D mkII in both ways (high light and shadow recovery): works well.

I just noticed when base ISO is high (like 5000 ISO) hot pixels are not fully registered, compared with a regular single RAW shoot at the same ISO.
But with a good de-rawtizer soft it's easy to set base ISO at 1600 and push to 5000 eq-ISO.

[a1ex]

I don't get it, but if you shoot above 1600 on 5D2 you are just throwing away useful data.

[NateVolk]

That's almost completely overexposed in one ISO. Can you reproduce it?

A video of the camera screen when metering the scene should help. None of my ETTR pictures in full sunlight looked like this...

I'll try again today with a video camera on the screen.  I re-downloaded the cr2hdr from the first post and it looks to be converting the missed files ok.  Thanks,

[yousufarain]

Hi,
I apologize in advance.  I am sure this has been asked ad nauseam but couldn't find it.  I am not sure what am I doing wrong here but I get an error when I drag the dng files to cr2hdr (mac user) saying files not supported.  I suspect I am not dialing in the right settings for the Dual ISO mode.  Can you please help. 

5D2 + Mac

- I the installed the latest build 4b3129d5a739
- Turned 'Disable all modules' to OFF
- Selected all the modules and rebooted
- Changed the Quality in Canon menu to JPEG (Does this need to be RAW?)
- Set RAW video ON (1880x1058)
- Set HDR to OFF
- Set Dual ISO to ON (100/1600)


I take the .RAW and convert to dng using RAW2DNG
I drag all the .dng files in a folder to CR2HDR but I get an error

Also the dng files look like regular RAW files and not the Dual ISO files like I expect to see them with interlaced ISO.

Completely new to this.  Thank you

[yousufarain]

Forgot to add that I am using the MAC GUI CR2HDR (Do I need something else too)

[akumiszcza]

Today I tested some dual iso + aettr. Notifications after settling ettr is great. Two things I've noticed:
1. In Av mode I sometimes get "ettr settled at iso 0/1600". I think it was when I had auto iso turned on in Av mode.
2. I get some photos without "dual" prefix in filename that are really dual iso. These files even have "Dual-ISO" keyword.

I would love to have an option in Aettr and dual iso to turn it off while not in M mode. Now if I have to take a quick shot without aettr and dual iso I have to turn these off in menus. Using aettr in Av even gives additional problems — if I turn auto iso, the option gets removed by auto ettr in work.

[akumiszcza]

Below is an example of a photo not possible without Dual-ISO. Not a very good one, but you can see both shadows (desks, closed doors), and very bright areas (in open doors). Canon EOS 50D, which is not good for low-light photos. Normally such shot require multiple exposure. In this case people would move and there would be quite a work in Photoshop. The problem is the fringing around people in the door. In my Photoshop CS5 there's no option to remove chromatic aberration in ACR (it appeared in CS6, process 2012), and I don't get good results in Lens Correction filter in Photoshop. I tried it in trial Photoshop CC with somewhat better results, but not great. I guess this is a "special" case of fringing, appearing only in Dual-ISO? Anyways, making near-HDR photos with one shot is a much greater advantage than some additional fringing

[jOt]

I'm still trying to rescue my shot, since not all of the frames converted. Cr2hdr failed on low key shots. Alex, would it be possible to create a version of cr2hdr without automated recognizing interlace, which extract just one iso channel, at half res, just to get rid of interlacing on unconvertable images?

[a1ex]

It requires the automatic recognizing to know which fields are bright and which are dark. This varies from camera to camera and sometimes from frame to frame.

Can you upload another CR2?

[jOt]

It requires the automatic recognizing to know which fields are bright and which are dark. This varies from camera to camera and sometimes from frame to frame.

Can you upload another CR2?

Thanks for explanation. In attached images fields are changing from frame to frame. Maybe a command of indication if the top field is bright or dark would rescue them?

http://www.mediafire.com/?20d3fr3hjrspb6n
http://www.mediafire.com/?j9lacm5cj4moohd
http://www.mediafire.com/?i4pmhpejwt61l5r
http://www.mediafire.com/?q8bkbh0vcq7n9he
http://www.mediafire.com/?exqi8cqvajlgkat

[1%]

Hmm.. I looked at one of those and there is so little data its probably confusing the algo. I guess it would have to check from the right where the barrel? is.

[a1ex]

I think the confusion comes from the black level difference: in dark areas, the low ISO pixels are brighter than the high ISO ones. The difference is usually small (say under 50 units), but it accumulates.

This should be the key to solving the autodetection.

A command-line option may help if you are willing to process each shot manually and fiddle with the offset until you get it right (I'm not). There are 4 values that you have to try.

[Marsu42]

A command-line option may help if you are willing to process each shot manually and fiddle with the offset until you get it right (I'm not). There are 4 values that you have to try.

It would be better than nothing - given the time cr2hdr takes to convert an image, batch-conversion with all 4 possible values wouldn't make much of a difference, at least to me if one of them has as little artifacts as possible.

If cr2hdr has any means to guess which of the 4 options is the correct one it would be nice to be able to append a "confidence" level or something like this to the file name so when sorting manually through them users would know which one is most likely to be the best.

Personally, I'm unlikely to use dual_iso all the time, so with the remaining images where I really needed it I'm prepared to spend a little time converting it.

[a1ex]

Only one of these 4 values will result in a valid image.

[Marsu42]

Only one of these 4 values will result in a valid image.

Ok, I didn't understand that - but imho one more reason to make them available from the command line if the (rare?) case should occur that the dual_iso autodetection chooses the wrong one? It's not like it'd be a big coding effort :-p

[a1ex]

It's not like it'd be a big coding effort :-p

In this case, feel free to submit a patch

[Marsu42]

In this case, feel free to submit a patch

Nonono, you got that wrong - it's not much of an effort for *you*, for *me* "Hello World" takes quite some time :-p

But to be earnest: I find submitting a full fledged pull request as it's desired around here to be too much overhead for trivial patches (long explanation cut, someone has to read this...) ... just posting a diff in a code box seems easier to me.

Also like in this case looking at some source code for the first time - and I'll never need to look at cr2hdr again - takes some time to get the bearings, just like working on the ml source takes some time if some months have passed. What I do in 1h you'll do in 10sec - I don't even remember how linux command lines are parsed and where the 4 options you mentioned might be located.

That's why I usually put my trivial or "open for discussion" ideas as a comment or request first, and only if push comes to shove change the source code myself. I know from dev's perspective this might look lazy, but just now I think for small matters this is a more economical way.

Speaking with Star Trek: I'm a doctor (no, actually I'm a sociologist) not a hardcore coder :-p

[jOt]

A command-line option may help if you are willing to process each shot manually and fiddle with the offset until you get it right (I'm not). There are 4 values that you have to try.

I'm begging for a patch. Everything would be better than cutting and pasting 2 pixel lines in photoshop on 50 frames . Please!

[Marsu42]

I'm begging for a patch. Everything would be better than cutting and pasting 2 pixel lines in photoshop on 50 frames

But it would teach you how to use photoshop actions :->