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

[dhilung]

Brilliant work!

Interesting black-level drift analysis and correction.

Btw, I also put a mirror at http://dhilung.com/vault/ml/dual_iso.pdf

[hetfanatic]

This would be very interesting to try out with my 7d. So far ML is amazing (even though it's not developed like 5Ds etc) and this would be so great to have. At least for fun. However, I'm not sure, since I'm not a developer of any sort, how to get this working on my 7D? Is it possible at this stage? If anyone can help, I'd be thankful.

[daisermac]

Very cinematic images - can't wait to see the first video footage. A1ex you are a genius - probably you will soon be abducted by some spy agency to work for their super secret decoding missions...

[jrumans]

Anxious to test this out, tried putting these files on my card but I know I am missing a step.  I guess we need to wait for Lourenco to compile for 5diii users?

https://bitbucket.org/hudson/magic-lantern/src/tip/modules/dual_iso

[Stedda]

Anxious to test this out, tried putting these files on my card but I know I am missing a step.  I guess we need to wait for Lourenco to compile for 5diii users?

https://bitbucket.org/hudson/magic-lantern/src/tip/modules/dual_iso

You're missing autoexec.bin

[jrumans]

You're missing autoexec.bin

Thanks for the heads up.

[Steve Kahn]

Would love to see a video of this!  Link?

[ml-images]

Hi,

Impressive work! One little question, I read in your pdf that the native ISOs are 100, 200 and so on, but some videos demonstrated that in video mode there is much less noise (aka native ISOs) on 160, 320, 640, ... at least on the 7D (growing noise order is 160, 320, 640, 100, 200, 400, 800, 1250, 125, 250, 500, 1000, 1600, 2500, 2000, 3200, 4000, 5000, 6400). Is there any logical explanation I'm missing here?

Regards, Michel

[Redrocks]

@ml images Were those videos talking about ML RAW or just plain old h264? I think A1ex mentions in his PDF that that does not apply here.


"As expected, intermediate ISOs like 160 or 250 do not cause any changes in ADTG/CMOS configuration.
These ISOs are obtained by applying some digital gain to the raw data acquired
at the nearest full-stop ISO, and this gain is configured from the DIGIC register 0xC0F08030
(SHAD GAIN). In LiveView, the gain is only applied to the YUV image (it does not affect the 14-bit
raw data at all), but in photo mode, the gain is burned into the raw data. Don't ask me why."

[ml-images]

@Redrocks indeed I'm talking Canon H264 here, not ML files. So my understanding is that Canon does change ADTG/CMOS configuration for 160, 320, 640 and doesn't for 100, 200, 400 which are obtained in software post processing. Does this sound correct?

[legreve]

I'm curious.... How much does the 5D mk III sensor differ from the sensor of fx the Alexa?

Could one force the data handling and interpretation from the Alexa down on the 5D chip?

[eatbuckshot]

@Redrocks indeed I'm talking Canon H264 here, not ML files. So my understanding is that Canon does change ADTG/CMOS configuration for 160, 320, 640 and doesn't for 100, 200, 400 which are obtained in software post processing. Does this sound correct?

I believe native ISOs refer to the available analog ISO gains to the sensor, which are the 100,200,etc. (set in the CMOS #0 register)

The ADTG/CMOS configuration can adjust only in those native increments, and the third stop gains are applied digitally after the nearest native ISO settings have been set in the configuration.

If I recall correctly, some videos that demonstrated the lower noise in 320, 640, etc were due to the fact that these are negative digital gain applied to the closest ISO of 400, 800, etc... which on our canon dslrs result in better shadow noise due to the fact that it is effectively "exposing to the right" then stopping down.  Even the PDF says that Canons are notorious for having bad shadow recovery.  This is why 125, 250, 500 yield higher noise since they are taking ISO 100,200,400 and digitally gaining +1/3 EV

[horshack]

It's interesting that Canon implemented this capability in their sensor. I believe the technique is patented by Eric Fossum, the inventor of the active-pixel CMOS sensor. Here is one reference to the idea on dpreview; there are other references but I couldn't find them quickly: http://www.dpreview.com/forums/post/32010565

[dadinio13]

after installing this we can remove it and return to the previous build ?

or the sensor settings are modified "for ever" ?

[togg]

Impressive. For real.

[Redrocks]

Most Online Today: 564. Most Online Ever: 564 (Today at 07:09:43 PM)

That's about twice the previous high and it looks like this has gotten everyone's attention. An old 4k thread got bumped the other day and I'd just been reading about how Arri are reluctant to move on from 2k, obviously to squeeze more money out of what they already produce, but also they feel that the current tech hasn't been fully exploited. Now this, The 5d3's shelf life just extended a few years.

[hjfilmspeed]

Hmmmmm half rez in shadows and highlights plus some moire. That might be ruff on video. Would be less of and issue on stills though. This is really interesting though. Amazing work!!!! Didnt even think this kind of thing was possible. I thought the same about raw vid from a dslr tho.

[payne]

Does it seem even remotely possible that this type of functionality could later be accessed for stills in APS-C lines that share a cut down version of the same sensor (if not the exact same sensor) as the 7D such as T2i and 60D?

[crazyrunner33]

Does it seem even remotely possible that this type of functionality could later be accessed for stills in APS-C lines that share a cut down version of the same sensor (if not the exact same sensor) as the 7D such as T2i and 60D?

Probably not; the 5D Mark III has an 8 channel readout along with the 7D, the others have a 4 channel readout from the sensor. 

[ted ramasola]

@a1ex,

would a 7D with VAF have minimal aliasing and moire?

[mesebar2]

You are a Genius.

[Audionut]

I can't get the windows executable to work.

dcraw is working fine, but cr2hdr reports "no such file or directory".

Code Select Expand

cr2hdr D46A8015.CR2

Input file     : D46A8015.CR2
'D46A8015.CR2': No such file or directory
Error: D46A8015.CR2

Wonder if it's related to this?

snprintf(dcraw_cmd, sizeof(dcraw_cmd), "dcraw -v -i -t 0 '%s' > tmp.txt", filename);

[jkdjedi]

 :)ANy Links to this MAgic?? Would like to try this cool hack.    (Canon 7D)

[Audionut]

I've got a 5D3 build here.

[jkdjedi]

Thanks Audionut, I wish I had the 5DMiii