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

[Magic 7D]

Wow A1ex! This is just getting better and better, so thankful for all your hard work. Noticed that the colours are a lot more vibrant with the latest version of cr2hdr too.

Now what I'm aobut to say is just dreaming.. 
Imagine if we could controll every second line vertically too, wouldn't this make it even more powerfull, or maybe better yet in a bayer pattern with 3 different ISO values like 100, 800, and 1600 just imagining it would bring even more DR to the final image. Now as I said it's just imagination on my side and I don't think it's possible. And maybe it wouldn't even help.. but then again who thought dual ISO would be possible before A1ex revealed it to us.

[Canon eos m]

Another update regarding aliased areas, for cr2hdr_exp.exe.

Hi Alex,

Could you please advise what is wrong with my cr2hdr_exp.exe post processing.

The original RAW Dual_ISO DNG shot with my 5DM3 is just fine.
https://www.dropbox.com/s/m86dyhnrbtajhi5/_G1A7637.CR2

Then when I processed it on the latest cr2hdr_exp.exe, I see big huge white patches all over the picture.
https://www.dropbox.com/s/rf44z0tqy4sl4ub/_G1A7637.DNG

[a1ex]

Ah, those tiny hot pixels.

It seems better with my latest copy (download the exe again), but there are still a bunch of them. I'll see what I can do, never seen that in my test samples.

[Canon eos m]

Grrreat!

Where can I get the latest version of ch2hdr.

Second, I downloaded enfuse after reading you used it for HDR type work. Any idea if enfuse will process CR2/
DNG files or will this require conversion first to TIFF/ JPEG?

[naturalsound]

Pixel peeping is pretty hard, who'd like to run some tests to make sure it handles pretty much everything better than the older one?

I just realized that the new version works great removing the reddish shadows. Unfortunately there are some artifacts near highlights which may be caused by the anti-aliasing? Each highlight is surrounded by a ring of horizontal lines which are about three to four pixels long.

I hosted the .cr2 at: http://www.naturalsound.de/MagicLantern/IMG_8485_ML.CR2
(I may have to remove the link if traffic gets huge).

Update: This CR2 is of an EOS 7D

The last experimental version did not show the lines and rendered the highligts clearly. Of course there was the reddish shadow cast.

I really appreciate your work!

[a1ex]

@Nikon eos m: this was ISO 12800/400? Here, the hot pixels are grouped, and my algorithm assumes they are isolated. Try not going higher than 3200.

To download, lookup my earlier posts (it's the same link). To merge with enfuse, I use this: www.magiclantern.fm/forum/index.php?topic=7022

naturalsound: here, the ISO response seems highly nonlinear, so the algorithm has trouble matching the brightness. Probably caused by aliasing, not yet sure.

[naturalsound]

Hi a1ex,

I forgot to mention the fact that this is an EOS 7D picture, but I guess you already noticed from the EXIF.
I have some additional RAW files in case you need another one.

Thank you for your fast response. I think it is a minor issue and will probably only happen in these very high DR situations. Maybe some pixels bleed because of the very high photon count and create the artifacts?

[Canon eos m]

Alex I call myself Canon eos m and not Nikon eos m - not that it matters, it is just that this a Canon forum and it is embarrassing as hell to have some one refer to me as a Nikon fellow :-)

[ted ramasola]

@a1ex


using aug 3 cr2hdr_exp


using aug 5 cr2hdr


using aug 7 cr2hdr


great improvement with color aliasing.

edit: used 7D

[a1ex]

Alex I call myself Canon eos m and not Nikon eos m

When using dual ISO it's like having a Nikon sensor

@ted: can you upload the CR2 for this shot?

[ted ramasola]

@a1ex,

compressed it and sent it to your email.

[Canon eos m]

@Nikon eos m: this was ISO 12800/400? Here, the hot pixels are grouped, and my algorithm assumes they are isolated. Try not going higher than 3200.

To download, lookup my earlier posts (it's the same link). To merge with enfuse, I use this: www.magiclantern.fm/forum/index.php?topic=7022

naturalsound: here, the ISO response seems highly nonlinear, so the algorithm has trouble matching the brightness. Probably caused by aliasing, not yet sure.

Alex got myself the latest cr2hdr.exe from page 1 of this thread. Like you said, magically, almost all hot pixels gone (at least the big and bothersome ones).

Same picture processed with the latest cr2hdr.exe
https://www.dropbox.com/s/vys1cux6yi7iege/kevin%27s%20workshop.JPG

[a1ex]

That's the old converter, which does not check whether the hot pixels are isolated or not. So it just corrects them blindly.

The problem is that it's too easy to mistake the hot pixels for aliasing. So, for now, I don't have a fix that solves both problems.

[Canon eos m]

That's the old converter, which does not check whether the hot pixels are isolated or not. So it just corrects them blindly.

Do you mean the one on page 1 is the 'old convertor'? If it is, where do I get the latest one?

[a1ex]

Another update regarding aliased areas, for cr2hdr_exp.exe

[AdamTheBoy]

Today's update keeps freezing as soon as it reaches the "looking for hot pixels" point.  Am I the only one with this issue? It seems to happen regardless of which image I try.  Thanks again A1ex for all the awesome work, it's exciting to see it get better and better.

[a1ex]

Can you upload a CR2? Shouldn't happen.

[Canon eos m]

That is the one I used the first time and ended with a lot of hot pixels. Will try to deploy it once again for another few pictures and see what happens! It is already 2:00 am here in Singapore. Good night!

BTW saw this awesome video today: http://nofilmschool.com/2013/07/magic-lantern-firmware-canon-linuxtag/

You folks rock!

[Canon eos m]

Alex, re-checked. It is only on this the photo I posted and only if I use the ch2hdr_exp.exe and not when I am using the ch2hdr.exe taken from the first page of this thread.

[AdamTheBoy]

Sure thing, here's one that I tried https://www.dropbox.com/s/4r0pfxn9kaepl87/7O4C8813.CR2

[Alex Roman]

I'm a bit confused. It seems that the last (7Aug) A1ex build has a huge step forward regarding antialiasing (which is really awesome BTW) but... how does this translates into video performance? Does it has huge advantage as well? Video samples anyone?

Thnx.

A1ex: i see, thank you. Looking so forward for it.

[a1ex]

It's not yet updated for video.

Updated the experimental cr2hdr again, this time fixed the brightness matching on 7D (test shot from naturalsound - the sensor response is linear, it was just my algorithm too sensitive to noise) and also fixed some buffer overflow errors.

I'm also on the way to fixing the black banding from 7D at ISO 100 (like this). How many of you did you notice it? Do you have more samples? Did you notice similar problems on other cameras?

[naturalsound]

Great work - Thank you!

I would not even have dreamed my 7D could deliver that clean shadows!

[IliasG]

I'm also on the way to fixing the black banding from 7D at ISO 100 (like this). How many of you did you notice it? Do you have more samples? Did you notice similar problems on other cameras?

I hope this helps a bit .. http://www.dpreview.com/forums/post/51945370

[Danne]

Maybe not the best test when it comes to dual iso but a little comparison between 3x cropmode and regular raw using a corrected 8mm peleng and a 20-35mm wideangle shot with dual iso.