Preventing Color/Luma Shifting When Processing DNGs in Adobe Camera RAW

[evoxio]

When processing DNGs from ML RAW video using Adobe Camera RAW (in either AE, Photoshop, or LR) the color/luma of the resulting footage shifts when the objects in the scene change. I'm a little surprised I haven't come across any discussion of this yet given how much of an issue it's been for me and how much I've noticed it in others' footage.

This can be seen subtly at 1:20 here (the luma of the water changes when the dark subjects cross the foreground):

Here's a shot where it's especially noticeable (watch the luma of the building in the background): https://docs.google.com/file/d/0B1keq-qyTGDNT21mR0hmWVFFc2s/edit?usp=sharing

This issue only occurs if you change the Camera RAW settings sliders (Clarity, Highlights, Vibrance, etc.) from their default/neutral states. It therefore seems like Camera RAW uses the color/luma values of the ENTIRE image to determine how to apply the demosaicing algorithm to any given pixel. In other words - the settings sliders aren't absolute controls, but rather they adapt based on the average color/luma of the frame. This phenomenon/method obviously wouldn't be an issue when processing still photos because the average color/luma of the frame doesn't change. But with video, it does.

Has anyone found a method to overcome this?

[squig]

I'm seeing sudden exposure shifts in pans in AE but it's not happening in Photoshop with the same footage. I think people haven't noticed it because they haven't been shooting a lot of movement. I haven't found any way around it in AE, I thought it was just a bug in the AE renderer.

[evoxio]

Hmmm... I just ran my DNG sequences through photoshop instead of after effects and the same phenomenon is occurring. I'm surprised that you're getting different results when using AE vs Photoshop, given that they both use the exact same Camera RAW algorithm.

Anyone else noticing this problem?

[haysuess]

I've seen this too. I have been using Adobe Camera Raw through Photoshop to process the DNGs. I've yet to try After Effects instead, but I'm thinking it'll be the same thing. This definitely makes a lot of footage unusable, so I hope there's a fix!

[mvejerslev]

I've seen this as well, a bit of exposure flickering visible in the portrait in

from 2:39. I'm not convinced its a Camera Raw problem. I know that the internal timer in cameras are not always 100% accurate, so exposures can shift (evident from years of panorama making in M mode). There must be a camera function in the h.264 encoding that prevents this exposure flickering happening.

[squig]

I'll run some more Photoshop tests today. Can everyone state which version of Adobe camera raw they're running, their OS, and their GPU.

[g3gg0]

This issue only occurs if you change the Camera RAW settings sliders (Clarity, Highlights, Vibrance, etc.) from their default/neutral states.

this proves that its not from the camera itself

[squig]

It's definitely not the camera.

[N/A]

This wouldn't have anything to do with using a linearized working color space, would it?

[squig]

Ok I've done some more photoshop + ACR tests and the exposure flickering is happening on any shots with movement/changing light where the highlight, contrast, etc sliders are used. I must have left the settings at default in my earlier test.

So essentially there's no way to make highlight and shadow detail adjustments to the DNGs with the currently available tools. Until somebody makes a tool like ACR that can properly process DNGs with movement we're screwed. We can't even create log style profiles.

[Levinson]

wow! I have shot a ton of raw on the 5d3 now, with plenty of pans, subject movement, exposure variations (shooting from 90 degrees to sun then panning into sun etc) and have yet to see this. I'm going to do some testing myself now.

I can certainly see it in the sample links provided.

Untill yesterday, I'd been using the 4gig limited Raw2dng converter and After Effects' ACR front end for all of the work.

I'm almost always reducing highlights, lifting shadows, adjusting white point and exposure in AE ACR- while occasionally adjusting contrast.

I always use ACR's "Auto" for white balance, or tweak slightly myself.

Using May 22nd 5d3 RAW video build.

Most footage I shoot is with full manual lenses like Samyang and Fujinon.

Sandisk 32gig Extreme CF (800x ??)

Shooting at 1600 x 900

It's possibly a sheer fluke that my combination of shooting environment and ACR tweaks don't reveal the issues.

[Andy600]

Does anyone know the exact RGB offset values for correcting the chroma shift that's happening? I can make a LUT that will give us a correct starting point for grading and then look at making the same with a LOG curve.

[squig]

wow! I have shot a ton of raw on the 5d3 now, with plenty of pans, subject movement, exposure variations (shooting from 90 degrees to sun then panning into sun etc) and have yet to see this.

I can certainly see it in the sample links provided.

If you're only making minor adjustments it may not be visible. Big adjustments like -80 on the highlights should break it.

[Levinson]

If you're only making minor adjustments it may not be visible. Big adjustments like -80 on the highlights should break it.

I understand what you are saying. However, I often almost always pull the highlights all the way to the extreme left combined with 80+ lift in shadows.

I use spot meter to check the % value of the brightest part of the shoot (including pans etc.) and adjust exposure to either 99 or 100% at that point. Spot metering has always been dead centre of Sensor.

Is it the the 5d3 you guys are using? The bulk of my recent work footage is on the 5D3. I can't say I've noticed it on 5D2 either, but have not shot as much with them recently (and haven't kept any RAW footage myself).

Sorry if the above seems fairly trivial. I thought that the more info I give the easier it is to isolate issues. I'll try changing various parameters in camera and see if I can reproduce it, even though we suspect it's RAW software related.

[squig]

MK3, but I'm 100% sure it's not the camera. It could however be a computer/GPU/ACR version issue. Can you fill me in on those details.

[Levinson]

yep will do as soon as I get to work.

[squig]

Does anyone know the exact RGB offset values for correcting the chroma shift that's happening? I can make a LUT that will give us a correct starting point for grading and then look at making the same with a LOG curve.

There's not much point making LUTs until this ACR issue is resolved. Have you tried rawtherapee.com It's a bit buggy but I played around with it a bit and you can apply LUTs and batch process.

[Levinson]

okay, quick yest here at home.

The clarity slider used above above 25 % starts to show symptoms of exposure changes(flutter or flicker). I rarely use it for video which might explain why I haven't seen it - but it's not as dramatic as what I see in the links.

Looks like shooting sun reflecting on rippling water might be a good test. Still with the settings used in previous post.

Dell L502x Laptop, i7 2670QM, Nvidia gt 540M, camera RAW 7.1, CS6, AE 11.0.1

Will keep checking

[squig]

Well we can rule out OS, GPU, and ACR builds. I think it's just that ACR just wasn't built with video in mind.

[Andy600]

@Squig - I haven't got Dynamic Link working with AE and ACR so AE just imports the DNGs into a wrapper and I can use colorista etc to pull back highlights just as much as ACR. It's just altering the shoulder for rolloff. If I know the chroma correction values I can make a LUT that simply corrects the RGB values and there would be no need to use ACR. I get no performance improvements using TIFF seqs over DNG seqs anyway.

[1%]

600D/6D and Rawanizer or ACR both showed expo shifts with fasts panning.

[eoshq]

I made a post about this back on may 19th, I don't think anyone noticed so I will say it again. http://www.magiclantern.fm/forum/index.php?topic=5404.msg37031#msg37031

As I mentioned before, beginning with ACR 7 and Lightroom 4, Adobe introduced what they call "Process 2012" which has automatic highlight recovery that it applies without user action. Merely loading a DNG, CR2, NEF, or other type of raw file will invoke this automatic highlight recovery that you have no control over, and which I believe has the potential to cause changes in rendering from one from frame to another because it was not designed for video. There is some discussion about process 2012 by still photographers way back in early 2012 when Lightroom 4 was in beta here. http://forums.adobe.com/thread/958989?start=0&tstart=0

I think you can avoid this by using "Process 2010" if anyone would like to test it out. I can't do it myself because I cannot run ML on my 5D3 until it supports firmware 1.21.

P.S. Please don't take this the wrong way but something I have noticed since the beginning of Magic Lantern is that there seems to be a huge disconnect between still shooters and video shooters/ML devs. Things that are common knowledge to raw still shooters, such as process versions in ACR, seem to be unknown in the video and Magic Lantern universe. I am sure there are other examples but I just find this divide interesting.

[squig]

I just tried Process 2010, it worked!    Danke.

Now we can make LUTs.

Ok so we're a bit slow, but we're working at 24fps, sometimes it's hard to keep up.

[1%]

Why is dcraw doing it too?


Not a problem for stills since they are shot 1x1.. or only a few at a time.

[haysuess]

I just tried Process 2010, it worked!    Danke.

Now we can make LUTs.

Ok so we're a bit slow, but we're working at 24fps, sometimes it's hard to keep up.

Thanks for the update squig! It's a shame because LR 4 clarity and highlight/shadows are so much better, but I guess it's a small price to pay.

Anyone know if it's possible to use the 2010 process in ACR via Photoshop, or can it only be done in Lightroom?