GH4 A fully logarithmic color space for Cinelike D and V-Log L

[Barry_Green]

Bad idea as exposure changes from profile to profile...significantly with V-LOG L.

Not having a GH4, I can't test, but -- is this necessarily the case? Because on the DVX200, the exposure is exactly the same for Rec 709 gamma and VLOG-L. The sensor needs a certain amount of light hitting it to do its work, and in both cases, the automatic exposure system chooses the exact same f-stop for both, and the resulting waveforms show that they're both exactly where they should be.

ETTR is tried and true

In stills, yes. In video, I find it more complex. ETTR chases the highlights at the expense of exposing your middle tones and darker tones all over the place, depending on how the highlights may have changed in the scene. Not a problem for stills, but can create significant hassles for someone trying to intercut footage from shot to shot, on shots that may have been significantly differently exposed. That's one reason why every log curve publishes the relevant middle gray exposure value, so you can get consistency from shot to shot; if you know where middle gray is supposed to be, you can expose Zone V consistently (whether that means spot-on, or if your preferences run to a stop over, or whatever; at least you can be consistent).

ETTR has its benefits, and I'm not ruling it out, I just find it -- tedious, to correct every single shot, just because some highlight might have changed here or there... then again, as always, I'm talking about VLOG on a DVX200, not on a GH4, where we have access to more exposure tools like the waveform, so it may not be relevant...

 

[balazer]

Exposure is a trade-off between noise and highlight range. To optimize the exposure is to minimize the noise while preserving as many highlights as you think you should. That's what you mean by ETTR. But which highlights are worth saving and which are not? There are specular highlights, lamps, objects under a spot light, backlit signs, LED & neon signs, reflective objects catching sunlight, bright sunny windows, etc, etc. The optimal exposure depends entirely on the scene and how you intend to process it. In some cases you will prioritize shadow range and some cases you will prioritize highlights. It is utterly meaningless to say ETTR unless you say which things should be exposed to the right. Just saying ETTR is saying "go shoot with the optimal exposure." It is bad advice for anyone who doesn't have the experience to know how to optimize the exposure across the full range of shooting conditions that he or she will encounter. For those with the experience, ETTR is fine, provided your workflow enables exposure compensation. And I surely do not mean pushing and pulling on the image with curves.

Optimizing the exposure is not always the goal. Sometimes you just want to reliably set a good exposure, even if it may not be the optimal one. Exposing with a monitor LUT, grey card, false color, or just shooting in a standard photo style with WYSIWYG monitoring are all more reliable ways of setting the exposure than ETTR.

I think enough has been said now about exposure. Please stick to the topic.

 

[joe12south]

Not having a GH4, I can't test, but -- is this necessarily the case? Because on the DVX200, the exposure is exactly the same for Rec 709 gamma and VLOG-L. The sensor needs a certain amount of light hitting it to do its work, and in both cases, the automatic exposure system chooses the exact same f-stop for both, and the resulting waveforms show that they're both exactly where they should be.

Sadly not with V-LOG L and the GH4 it's at least a stop different.

In stills, yes. In video, I find it more complex. ETTR chases the highlights at the expense of exposing your middle tones and darker tones all over the place, depending on how the highlights may have changed in the scene. Not a problem for stills, but can create significant hassles for someone trying to intercut footage from shot to shot, on shots that may have been significantly differently exposed. That's one reason why every log curve publishes the relevant middle gray exposure value, so you can get consistency from shot to shot; if you know where middle gray is supposed to be, you can expose Zone V consistently (whether that means spot-on, or if your preferences run to a stop over, or whatever; at least you can be consistent).

ETTR has its benefits, and I'm not ruling it out, I just find it -- tedious, to correct every single shot, just because some highlight might have changed here or there... then again, as always, I'm talking about VLOG on a DVX200, not on a GH4, where we have access to more exposure tools like the waveform, so it may not be relevant...

Definitely some valid points, and if I have the opportunity to "set-up" shots I love to have a grey card to work from, because as you said, it makes post much easier. But oft times it's not available or realistic. If I'm chasing a bridal party around a park, I don't have that luxury. When you have to expose by the seat of your pants, ETTR is the way to go on the GH4, because, again, underexposure is just horrible.

 

[joe12south]

I think enough has been said now about exposure. Please stick to the topic.

Okay, well I think I'm out on this topic. I can't find any practical benefit to your LUT, just a lot more extra work to get to a usable image, and you don't seem to be interested in real-world application versus theory.

 

[balazer]

My LUTs are all about making basic color correction easy, in the same way they're easy for raw images. I did my best to try to explain that and to help you with the workflow.

 

[royfel]

I produce for PBS and use the GH4, lately editing totally in Resolve. I am shooting UHD 29.97 then creating a ProRez 1080 444 file using Cliptoolz as an intermediate. I have shot a couple of assignments using the balazer method and am extremely pleased (placing the CineD to log converter as a preset in Resolve) I have tried just about every setting (including Lemming) and find the workflow mentioned here to be the most pleasing and easiest to expose for in the field. (especially the highlight rolloff LUT). I am not inclined to shoot tests other than for myself just saying that this workflow provides a very solid and well thought out approach to data wrangling and in my opinion only gives the best I have been able to produce from this camera.

 

[balazer]

Thanks, royfel. Do you have any workflow tips?


I have published a new LUT package, v0.12, now with calibrated color for the V-Log L output of the GH4. Before I was using the Panasonic V-Log/V-gamut transform. Calibration has improved the color accuracy of V-Log L, especially in blues and greens, which were oversaturated. V-Log L can now be matched more closely to Cinelike D.

Cinelike D: (click for full size)


V-Log L with calibration:


V-Log L using Panasonic's V-Log/V-Gamut transform (no calibration):



The calibration is not perfect, but it's a lot better than Panasonic's transform, and about the best I could do with the images I have. It's a bit surprising just how far off V-Log L is from V-Log.

I plan to do the same kind of calibration for Cinelike D.

 

[CustomerGH4]

Just got a call from a store where I got GH4, they talked with Panasonic and turns out it is not available for my country.
So I guess the no further updates for me if I want to keep the log.

 

[mash_gh4]

..., now with calibrated color for the V-Log L output of the GH4. Before I was using the Panasonic V-Log/V-gamut transform. Calibration has improved the color accuracy of V-Log L, especially in blues and reds, which were oversaturated. V-Log L can now be matched more closely to Cinelike D.

i have to agree. panasonics gh4 implementation of v-log doesn't comply to the specification giving by them very well. there are obvious deviations. i first saw it on in this gamma-graph in the panasonic presentation when they announced v-log. exactly the same differences i found in all my own testings.

in the meanwhile i have worked on my own tentative ACES IDTs for all native gh4 picture profiles. i also stumbled over this horrible [over-]saturation problems and still could not figure out a sufficient way to handle the color correction. on the base of this efforts an a lot of measured test charts and transformations i feel quite skeptical, if a linear compensation of this color errors is possible anyway? there are some very advanced research papers available, how to solve this challenge by new strategies, but it doesn't work very well using more traditional tools.

 

[balazer]

Yes, V-Log L also has deviations from the V-Log gamma curve. The GH4 is putting black too low, creating a picture with a bit too much contrast. Mash, if you want to provide me with your 1/3-stop exposure series for V-Log L, I could compensate.

 

[Lpowell]

I have published a new LUT package, v0.12, now with calibrated color for the V-Log L output of the GH4. Before I was using the Panasonic V-Log/V-gamut transform. Calibration has improved the color accuracy of V-Log L, especially in blues and reds, which were oversaturated. V-Log L can now be matched more closely to Cinelike D.

Which gamut do you think V-Log-L on the GH4 is actually using?

 

[balazer]

I don't know. It doesn't look like any named gamut that I know of.

I mispoke. It's mostly the blues and greens that were oversaturated.

 

[balazer]

I have posted v0.13 of my LUTs, with these changes:

• DaVinci Cube format added
• support for Cinelike D 16-235 and 16-255 added
• support for video applications that don't read the video_full_range flag added
• xvYCC decoding of Cinelike D added
• test images and test LUTs added

Please let me know if you see any problems.

 

[mash_gh4]

Which gamut do you think V-Log-L on the GH4 is actually using?

it seems to use v-gamut as described here: http://pro-av.panasonic.net/en/varicam/common/pdf/VARICAM_V-Log_V-Gamut.pdf

 

[balazer]

No, mash, it's not V-Gamut. That's what I tried first. When you interpret V-Log L's colors as V-Gamut, they come out too saturated.

 

[mash_gh4]

No, mash, it's not V-Gamut. That's what I tried first. When you interpret V-Log L's colors as V-Gamut, they come out too saturated.

did you use one of matrix transformations from the quoted specification or did you calculate your own formula?

i get acceptable results interpreting the gh4 v-log test shots with my own linearization and panasonics v-gamut color transformation matrix.

but i failed until now to reconstruct this transformation matrix from scratch -- especially the included chromatic adaptation...

 

[FreakPhoto]

Error in Color Finale/FCPX trying to use your DaVinci LUT:
"Error in applying LUT - Data length in file does not match declared size"
Normally Color Finale loads DaVinci LUTs without error.

 

[balazer]

Thanks to FreakPhoto for running a bunch of FCPX tests for me.

FCPX update:

• Color Finale is able to load Adobe Cube 3D LUTs correctly, but not DaVinci Cube LUTs.
• LUT Loader is able to load DaVinci Cube 3D LUTs correctly, but not Adobe Cube LUTs.
• Neither plugin can load 1D LUTs. My log-to-gamma LUT is a 1D LUT. I will work on a 3D version of that transform.
• I still don't know if FCPX maps video levels in a way that the LUTs expect. The instructions in the readme.txt file should help you test whether levels are read in the expected way by FCPX and mapped correctly by the LUT.

Mash, I took some V-Log L footage, applied the V-Log-to-linear transform and the V-Gamut to XYZ transform, and calculated the chromaticities of some color patches. I then calculated the chromaticities of the same patches in a GH4 photo of the same target under the same illuminant processed from raw. I used the differences between the two sets of chromaticities to calculate a correction matrix, moving the colors of those patches from where they were decoded from V-Log L to where they should be based on the processed raw photo. This form of correction is based on the assumption that the color accuracy errors are due to an incorrect raw-to-V-Gamut matrix inside the GH4. It's not a guaranteed accurate form of correction. It suffers from measurement error, and it will suffer from whatever color errors the raw processing suffers from. But at least it makes V-Log L colors look more like a raw photo's colors, and I think we can agree that processed raw GH4 photos have better color accuracy than any of the GH4's video modes. It also avoids the complication of knowing the exact spectral reflectances of the color patches and the illuminant, since the V-Log L video and the raw photo were shot in the same time and place of the same target.

 

[Dr_Jon]

A quick question - if all that was from my test data did you compare the Canon Raw as well?

 

[royfel]

Over the weekend I did an assignment in very difficult lighting with over 100 clips for PBS (shot Cine D as suggested). I put the 'GH4 to log' LUT into Resolve preferences>color Management> input 3d LUT and then created a timeline node for the highlight roll off LUT. I shot a color chart and used Resolves' "chart correction" feature as a separate timeline node after the high light roll off. I was one of the first people to get this camera and so far this is the best footage I have ever seen from it. Skin tones perfect under really bad mixed gymnasium lighting. I used the new Resolve LUTS for everything. Sounds time consuming but it is very fast if you do it the way I did. (Tip: unless you have the mother of all computers take advantage of "generate optimized media" when working in UHD.)
Thank you so much, it's like owning a new camera.