Author Topic: Magic Lantern Cinema Camera - Dual ISO without aliasing & without quality loss!  (Read 79512 times)

squig

  • Hero Member
  • *****
  • Posts: 518
  • Crash test dummy MK3
Cool thx. Yeah it's beer o'clock here too :D

800/100 ISO 1x3 10bit



same shot +1 EV




Lars Steenhoff

  • Senior
  • ****
  • Posts: 473
Cool results, and yes I see the stripes in the shadows

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
If entering canon menu and back before filming stripes are often gone. Except on the right hand of the image. Well, more tests...

flostro

  • New to the forum
  • *
  • Posts: 38
  • EOS 5D Mark III / EOS 3 / A1

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 8853
If you don't know a version you will find a <cam identifier>_<firmware version>.sym file in ML\Modules. Here it is 5D3_113.sym.

flostro

  • New to the forum
  • *
  • Posts: 38
  • EOS 5D Mark III / EOS 3 / A1
Good to know, thank you

70MM13

  • Hero Member
  • *****
  • Posts: 607
I tried your 14 bit version and it works very well.  Straightforward and easy!

Unfortunately (or fortunately, depending on your point of view) it doesn't seem to provide any benefits compared to the ISO experiments, for me at least...

I will keep watching and playing with any new developments you make, and hopefully you will work some magic!

I think that dual ISO combined with the ISO experiments is the way to get truly incredible results...

If only more people were playing with the ISO experiments, maybe this could happen.

I'll try some more tests with your 14 bit version and see if it beats the ISO experiments in some cases.  The shadows are just so clean in the ISO experiments...

I hope you don't think I'm discouraging you.  Quite the opposite!  Keep up the amazing work!


70MM13

  • Hero Member
  • *****
  • Posts: 607


Much better!

But something strange is happening now.  When I load a clip from the "less stripes" version into mlvapp, as soon as I switch to anything except bilinear demosaicing, the brights lose the dual ISO processing.

I checked by reverting to the prior version and it's fine.

I'm using the 14 bit version.

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
Seems problematic still. Think it's fps related. As soon as I lower fps to something like 22 stripes are gone. Finetuning I guess.

70MM13

  • Hero Member
  • *****
  • Posts: 607
It was the shutter speed.  Seems like it was somehow set in your version.

Once I set it to 50 in the ml exposure menu, the problem disappeared!

For some reason, when I set FPS override, the camera crashed hard. Display stays on even with the battery door open.  Battery has to be pulled.

This version provides very nice results!  Maybe better than ISO experiments!

I'll do a comparison with this.

EDIT: I can't repeat this.  It must have been a stray neutrino.  But the hard crashing is definitely repeatable :p

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
Uploaded two new versions. Finetuned, hopefully stripe free.
And yes, stay off fps override...

Levas

  • Contributor
  • Hero Member
  • *****
  • Posts: 1743
  • 6d - Nightly build user
FPS override doesn’t combine well with crop_rec presets.
FPS override changes A and B timers, which are already carefully set in the crop_rec preset.
This combination probably also messes up shuttertime, as calculations are done with the settingsfrom the crop_preset.

70MM13

  • Hero Member
  • *****
  • Posts: 607
The new version fixed the interlacing issue, but the vertical stripes are very strong, especially if underexposing for highlights.

I did many exposures of this test scene, and even when I exposed exactly for midtones, the stripes are visible on the wall well into the mids...

As for dynamic range, it doesn't beat the ISO experiments yet.  Pretty much the same.

But, if you can eliminate the stripes, it could be a winner!

I still dream of both combined.

The following images are straight from mlvapp with no processing except exposure and shadow strength to match.  No noise reduction nor chroma smoothing applied.

The ISO 109 was shot at 3K and exported at 1080.  BTW, the 3K footage uses less data...

Different framing due to different crop sizes.




Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
I just posted two new versions. Only tested 10bit briefly. Not sure why you still get stripes. Maybe unavoidable. Thanks for testing. Feel free to test these two latest ones as well if you have the time. Thanks.

70MM13

  • Hero Member
  • *****
  • Posts: 607
I tried the new version and it's the same.

I hope you can solve this!

Simple test: expose for a light bulb in a dimly lit room just like the scene I did.  You'll know right away!

theBilalFakhouri

  • Developer
  • Hero Member
  • *****
  • Posts: 1346
  • UHS-I
In 700D there are no problems when using 10-bit lossless analog gain possibly also other small cameras like EOS M, 650D and 100D will be not affected. Maybe the cameras which has ADTG4 analog registers will be a little bit tricky like 5D3.

Canon 700D:
Original" border="0

Pushed-5-stops" border="0

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
I tried the new version and it's the same.

I hope you can solve this!

Simple test: expose for a light bulb in a dimly lit room just like the scene I did.  You'll know right away!

Could you show me an example of the problem with the 10bit build please.

70MM13

  • Hero Member
  • *****
  • Posts: 607
10 bit version:


Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
Thanks. Will try and reproduce. And fix maybe.

70MM13

  • Hero Member
  • *****
  • Posts: 607
Good luck!
I hope you succeed, I'm starting to like this!

12georgiadis

  • Member
  • ***
  • Posts: 213
  • 5DmkIII - 7D - EOS-M
Hi! If there is a build with eosm dual iso lossless 10 bits, I can do some tests


Envoyé de mon iPhone en utilisant Tapatalk

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7740
Uploaded two new builds.
1920x2320, fps23.976

Tried 10bit in daylight both dualiso and regular raw and footage looks ok. Tried some stress testing against a lamp but couldn't find any stripes.
Feel free to do some stress testing against either build.

IDA_ML

  • Hero Member
  • *****
  • Posts: 1014
Hi! If there is a build with eosm dual iso lossless 10 bits, I can do some tests

Same question but about the 100D.  Can we see a working build on that camera?  I can't wait to test.

squig

  • Hero Member
  • *****
  • Posts: 518
  • Crash test dummy MK3
Uploaded two new builds.
1920x2320, fps23.976

Tried 10bit in daylight both dualiso and regular raw and footage looks ok. Tried some stress testing against a lamp but couldn't find any stripes.
Feel free to do some stress testing against either build.

Stripes still there on the 14bit build.