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 5 Guests are viewing this topic.

kyrobb

Any discoveries recently that could lead to dual ISO working in 50D and 5D2 video?

N/A

QuoteDownload cr2hdr-20bit for Mac only with last commit 0eabcb0 ;)

Download here
I get the following exiftool error when using this build of cr2hdr:
Hexadecimal number > 0xffffffff non-portable at /usr/bin/lib/Image/ExifTool/Writer.pl line 2878.
Hexadecimal number > 0xffffffff non-portable at /usr/bin/lib/Image/ExifTool/Writer.pl line 2878.


They open fine in ACR/LR but have no preview, so if I use the --compress option, which enables the .dng to be displayed in preview, it comes out like this-
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?

kichetof


ayshih

Quote from: kyrobb on September 27, 2014, 07:37:22 PM
Any discoveries recently that could lead to dual ISO working in 50D and 5D2 video?
Nope.

Quote from: N/A on September 29, 2014, 05:13:43 PM
They open fine in ACR/LR but have no preview, so if I use the --compress option, which enables the .dng to be displayed in preview, it comes out like this-
If you mean OS X Preview, there's a known issue where Preview apparently ignores the EXIF black level if it recognizes the camera model (and presumably uses some default black level), which can lead to color casts like you show.
Canon EOS 50D | 17–40mm f/4L & 70–300mm f/4.5–5.6 DO IS | Lexar 1066x

N/A

Yup, exiftool 9.72. Looks the same in Xee, not sure about any other image viewers. Figured it was a black level issue, if its a known bug then that explains it.
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?

swinxx

hello a1ex,
i have a problem with a specific shot, perhaps you can take a quick look at it.
the shot is made with a canon 5dmk3 and there were no extreme settings applied.
after conversion the file gets some pinkish cast where the wood window fragments nearly touch each other.

here is the link to the dng: https://copy.com/3k6ep8dx7D1B3aNm
one original, converted with mlvfs, and the other one is converted with the lightroom plugin.


can i save that shot and get rid of that pinkish edges?
thank you

a1ex

That's aliasing. I'm not sure what I can do about it, maybe identify high-contrast areas and desaturate them?

swinxx

Hello A1ex!

Thank you, that sounds great, cause the problematic area is pinkish :)

another interesting observation is, that after conversion from untouched to converted dual iso dngs with lr plugin  - in a frame pattern of A B A B A B A B.... after bringin up the shadows my video has
A.. greenish blacks and
B.. magenta blacks
in Davinci Resolve,

when i am at home i will post 2 frames where you can see that.?
is that a side effect of a too dark exposure?

thx swinxx

ayshih

See related discussion.  The white-balance calculation in cr2hdr is apparently being affected by the fact that which lines are at which ISO changes from frame to frame in dual-ISO video (in crop mode?), which is probably changing stuff like black levels.
Canon EOS 50D | 17–40mm f/4L & 70–300mm f/4.5–5.6 DO IS | Lexar 1066x

a1ex

There is another effect in video mode: at certain FPS settings, one frame ends up as dark-dark-bright-bright, the next one is dark-bright-bright-dark and so on.

And since the RAW video does not contain optical black data (for speed reasons), the black level is guessed. The guess may be different in the two cases (which may cause green/magenta difference in shadows).

The DNGs will tell exactly what's going on.


Danne

Here is Senzazn two DNGs with whitebalance shifts.

From here
http://www.magiclantern.fm/forum/index.php?topic=12796.msg130197#msg130197

"Camera is a 5D Mark III. I used the CR2HDR20BIT algo through Magic Lantern RAW Video Converter 1.9.2 to process the MLV files to Dual Iso DNGs. Then when I opened up the DNGs in Lightroom, despite syncing them all to have the same white balance, each alternating DNG frame has a slightly different white balance. Here are two sample DNGs.

https://www.dropbox.com/s/kimafdub1vm9pm0/MLV_A002_271841_C00016.000000.dng?dl=0

https://www.dropbox.com/s/c4ac4t5qr5usy1w/MLV_A002_271841_C00016.000001.dng?dl=0"

SpcCb

Audionut > Nice! ;)

Edit: the CR2 looks... Strange (?); 19.1Mo, with no header informations.

dark-shadow

Not sure if this is the right place to ask, but since cr2hdr belongs to the dual_iso module and I found this thread later, maybe I try and ask if someone can help me with my problem describe here?

http://www.magiclantern.fm/forum/index.php?topic=13548.0

Audionut

Quote from: SpcCb on October 08, 2014, 07:00:42 PM
Edit: the CR2 looks... Strange (?); 19.1Mo, with no header informations.

I guess Canons lossless compression works well on these images.  A quick scan through my collection sees another dual_iso image @ 100/1600: 28.1Mo.
Here is an non dual_iso image from the intervalometer sequence, five shots before the above:  https://dl.dropboxusercontent.com/u/34113196/ML/dual_iso/_MG_5684.CR2

Not sure what you mean on the header info, all the exif data looks intact here, minus focal length/aperture (telescope).




An issue I found, when using the intervalometer, enabling/disabling dual_iso would not work for the next captured image, it was the next image after, that dual_iso became enabled/disabled.

SpcCb

Fault on my side; CRC error during download... (I use a crap VPN)
Re-download and it's OK. ;)

BTW I'm surprise of the size of the 6D file; with 5D2 'Moon' Dual ISO are very heavy, something like 30~40Mo, more than an regular CR2.
Maybe a question of noise... Or optical sampling/focus.

Audionut

Resolution differences aside, the 6D is one of the best Canon performers regarding noise:  http://www.clarkvision.com/articles/evaluation-canon-6d/index.html
I downgraded from a 5D3 to 6D for the noise performance, as I'm looking to move more into astro photography.

I also use ADTG tweaks  :)

I used an NexStar 6 SE (cheap, all things considered), haven't made an equatorial mount yet, and missed prime focus (seeing conditions were poor), so there were no fine details to boost filesize.

256256256

The dual iso have something which is similar to interlaced video, in the interlaced video lines alternates in different fields with time difference, in dual iso lines alternate with light difference but no time difference. No matter this comparison, both interlaced video and dual iso video suffer from aliasing after separating the two images. The link below shows an excellent result for deinterlacing interlaced video, maybe this solution or a modified version can do a good job for dual iso:

http://forum.doom9.org/showthread.php?t=156028

zigazaga

Quote... 5dmkiii ... ... developments ... coming forward on this platform.
<- I'm glad this has been clearly stated!

I think ML should put this type of information very prominently on the home page - e.g. "Want to be on the edge - get a 5D3!!!" kind of thing. And do not be embarrassed to monetise the promotion - it truly would appear to be the best platform for advanced techniques, right?

I also believe there should be a user-friendly website built to help the simple photographer navigate and select the features that may be relevant to their method and their gear. I can see a clear potential for monetisation there too - e.g. a site that prompts to create and install a personalised custom build tailored to a registered user's profile. I personally only want a few useful features and rather not get confused by the choice between various semi-functional builds vs the latest "experimental bells and whistles" version irrelevant to my camera model and my requirements. E.g. I'd rather pay to get a simplified yet customised build that does no more than what I want and *not* have to study any esoteric complexities.

I believe that the majority of less technically minded photogs will be prepared to pay the price of a memory card to subscribe to such a personalised service as they may be intimidated by the complexity or lack of features in the full builds - the 2 year old one that's currently presented as the safe one vs the latest more complicated one. Think about it - a services that charges more for fewer but relevant features rather than trying to be an "All things for all men at all times" type of thing!

ap1hk

Hi a1ex,

I have been using ML for more than a year and am using dual ISO for roughly half of the photo shots.
I would like to suggest that you can add a flag to cr2hdr to force export DNG even it is not an interlaced ISO CR2 file. It will help a lot for people like me who is not using dual ISO all the time to standardise the workflow (working with DNG alone). I see your code that there is a debug routine to output the file untouched which is actually what it will do.

Audionut

Identical settings.
ADTG[888x] 32x
Looks to me like the dual_iso exposure has the most accurate feedback.  I guess this is because you have accurate white level from the higher ISO?




https://dl.dropboxusercontent.com/u/34113196/ML/dual_iso/_MG_6014.CR2
https://dl.dropboxusercontent.com/u/34113196/ML/dual_iso/_MG_6015.CR2

a1ex

The second one should be marked as blown out, right? It's a white level problem (not detected correctly if you decrease the gain too much).

Audionut

Zebras go weird at 31x, and white level drops at 30x.

RawDigger reports the whitelevel of the non dual_iso image as 14286, and since I still have aperture gains active, saturation should be 16283 (f/1.4).  So it looks to me, the second one (dual_iso) has the correct exposure feedback, and the first image (non dual_iso) is incorrect.

I never noticed this behaviour on 5D3 (different exposure feedback for dual_iso/non dual_iso), and I took tens of thousands of images on that with various gains, and dual_iso.

a1ex

Possible. I should be able to debug it in QEMU (will convert the CR2 to DNG and load it there).

DeafEyeJedi

This is quite interesting... definitely makes sense with the fact that the dual-ISO giving correct exposure feedbacks because its actually reading it 'deeper' within both ISO's as oppose to just one.

5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109