Informative shimmering in dual ISO?

Started by 70MM13, December 06, 2017, 01:08:12 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

70MM13

Hi everyone...

I've got some interesting footage I shot the other day for a new film I'm creating.  I'm really digging the benefits of dual iso, but sometimes strange things can happen.  In this example, it seems as though the interlacing is alternating in the brightly lit patch of clouds above the sun.

Luckily, I'm shooting each clip in both single and dual iso modes in case of such things, but I want to share this example in case it's unusual enough to hint at something to investigate.

Shot on my 5Diii, firmware 113 using the latest nightly build.

The frames were extracted using the very new build of mlv_dump.

Every combination of ways to play with the frames on my windose machine has been tested, and the shimmering is always there.

It may be quite normal to you, but somehow it seems a little odd to me!

I'd upload the mlv but it's far too large, so this handful of frames should suffice...

Thanks for this incredible enhancement to so many photographer's lives!

https://drive.google.com/file/d/17jD44hIJQLsAtkrtyInK1_Xa3f2vae31/view?usp=sharing

Danne

Not sure what you mean but you don´t mention cr2hdr so maybe we can assume you didn´t process the dng files after extracting with mlv_dump?
Here´s two versions of cr2hdr:
http://www.magiclantern.fm/forum/index.php?topic=7139.msg194062#msg194062

a1ex

We used to call this effect "crawling" - you can avoid it, but that won't reduce the aliasing.

http://www.magiclantern.fm/forum/index.php?topic=9456.0

It's very hard to fix, except in very simple cases like this, with little or no motion (where it's very easy to fix). Will post a script in the main Dual ISO thread.

70MM13

Thanks to both of you for the replies!

I'm looking forward to playing with the script you will provide. Hopefully it rescues this otherwise excellent shot!

While I have your attention, please accept my humble request to continue with the "low" ISO experiment. Anything to help clean up shadows is VERY eagerly desired!

I'm sure that I'm not the only artist who's wishing for it.  I'll help with real world testing as much as possible but I cannot delve into compiling and all that.  Give me a build and I'll torture test it!

Thank you!

a1ex

FYI, the current findings (ISO 66 etc) are already available for 5D3 in the iso_regs module (check the download page, no need to compile anything). They are just not very user-friendly (settings not saved, each ISO needs fine-tuning), but people can already experiment with these settings and provide feedback. Without compiling.

Of course, they are still on my list.

70MM13

Fantastic, thank you!
I'll start playing with it as soon as I'm done installing octave and applying your fix to the affected shot.

One more question for you:

Is it possible to artificially enlarge the dng files by 4 sensels on each image boundary?

The reason I ask is because I'm very comfortable with doing post using rawtherapee, and it is hopelessly truncating dng files by that amount.  I've raised the issue with them and it seems that it will not be resolved.

I'm a fan of 1080 raw with 3*3 binning, and unless I'm mistaken there's no way to increase the recording resolution in that mode, particularly since I'm doing h264 proxies and want the proxy and raw to be identical resolution...

Any advice would be greatly appreciated!

a1ex

Of course, just fill with zeros or (maybe better) with the boundary values from each Bayer channel. See e.g. padarray:

https://octave.sourceforge.io/image/function/padarray.html

This is a very good guide if you want to get started: https://rcsumner.net/raw_guide/RAWguide.pdf

(I might post a tutorial on this as well if others are interested)

70MM13

Thanks again!
I'll be quite busy playing with all of this.
I'm tempted to set aside my dual iso shots and start playing with iso66!
I'll definitely share the results.

Regarding the dng padding, an alternative solution for the rawtherapee problem could be to convert the dngs to cr2s, if it's possible?

I'm looking for the least troublesome workaround to a problem that shouldn't exist...

Thanks again for your replies!

70MM13

OK, here's the results of a very unscientific first attempt to play with the registers:

Both are frame grabs from the corresponding mlvs, all camera settings identical except the "iso".

The images are taken from rawtherapee, with only debayering and chroma noise reduction applied.  Luma is untouched.

Note that although the highlights are pinkish, there is more highlight detail in the iso68 shot, particularly the threads on the light bulb socket!

Also note that although fewer photons were gathered, there's definitely a bit less noise in the shadows. Look closely at the inner edge of the speaker cone just outside the dual concentric HF driver.

Not a huge difference, but ANY improvement is worthwhile!

I'll keep happily experimenting...