GH4 Exposing V-LOG L

[Barry_Green]

With Canon C-Log, highlight saturation is at 109 IRE and with Panasonic V-Log-L, highlight saturation is at 79 IRE and we can all see what a difference that makes.

With all due respect -- I don't think you can. The difference between the 79 and 109 IRE cutoffs does not explain the purple/green in the lower 9 stops of the gamma curve, where both CLOG and VLOG-L encode fundamentally identically.

Absolutely regardless of Canon's marketing, the fundamental fact is that both CLOG and VLOG-L record 12 stops. They record the first 8 stops fundamentally identically. They record the last 4 stops differently, with the Canon using more codes. Do you disagree with any of this? I would be shocked if you do, but I do have to ask to see if we're at least on the same page to this point.

The brightest four stops have more gradation in CLOG than they do in VLOG-L, that is true. It is also where it matters the least; it only comes into play on items that are hotter than 90% white reflectance. I would further say that much of the purple/green we've been seeing demonstrated is not even happening in those upper stops; most certainly the stuff on the road is not above 90% white reflectance!

The problems in the GH4's purple/green stuff are not attributable to VLOG-L having fewer codes. This is largely definitively proven by the DVX200 rendering the fundamentally same scene, using VLOG-L, in the same 8-bit space, in the same codec, with the same 79-IRE cutoff, without any of the purple/green stuff happening. It is not anything to do with VLOG-L's gradation steps. It just isn't. You may not LIKE the idea that VLOG-L has fewer steps, but that disapproval does not convict it of this crime. The gradation steps in VLOG-L are sufficient to avoid this problem, as the DVX200 shows. Whatever the problem is, it isn't the VLOG-L code values or 79 IRE cutoff.

Where is this horrible macro blocking coming from? Is it the excellent Panasonic codec ruining the party?

I don't have definitive answers for you as I don't even have a GH4 to experiment with. I can tell you some places that it's not coming from, and I can speculate where it is coming from.

The first place it is coming from is in the post-processing chain, as we've already seen proven with Premiere. Some programs are causing blocking that doesn't show up when other programs are used. Those who are encountering the issue should coordinate to definitively determine what post-processing software handles VLOG-L recordings poorly.

It is my contention that, as others have already said and shown, the purple/green is always there in all profiles, but it is there to a much lesser degree in the other profiles. It shows up on VLOG-L more because, I theorize, the color saturation on VLOG-L is so much lower to begin with, that it requires more exaggeration in post to return to proper saturation. When doing so, the purple/green get magnified as well, and what was once a hidden little issue becomes more to the forefront.

Can you turn down the color in a GH4 sufficiently in any other profile that it matches the color saturation of VLOG-L as recorded natively? If so, someone should try running that test to see if a desaturated regular profile exhibits comparable purple/green to what you're encountering with VLOG-L.

Secondly, I would invite someone to communicate with the beta testers who've been using VLOG-L on the GH4 for the better part of a year; it would seem like those with the most experience would have run into all these problems and already have worked out solutions to them.

 

[balazer]

The fact is that it's compliant with VLOG, and so those bits are only used in sensors that deliver more than 12 stops, and the GH4 doesn't, so those bits don't get used. In order to utilize those bits, they'd have to create a new gamma curve, a not-VLOG gamma curve.

Not true. At higher ISO settings than the base ISO of 400, a GH4 with V-Log L could very well use a greater range of color values and still be V-Log compliant. That's how Canon Log does it. They put their log base exposure index at a higher ISO setting than the minimum ISO setting in log mode. The fact that Panasonic didn't shows they really don't understand log color spaces.

But absolutely, Panasonic also missed an opportunity to create a whole new log color space matched to the dynamic range of the GH4 and optimized for 8-bit recording. They would have had to call it something new, something like, I dunno, V-Log Lite.

 

[balazer]

But -- I'm still not sold that this is an 8-bit problem. It just isn't. Can't be. People have been using C-LOG for 3 years in AVCHD, 5 years in MPEG-2 8-bit, and nobody's complained about anything like this.

It's not just an 8-bit problem. It's the problem of 8 bits being used in a color space with a very wide dynamic range and a large color gamut. V-Log was designed for 10 bits from day 1, and it doesn't make particularly efficient use of code values. Canon Log was optimized for 8 bits. It has a smaller dynamic range and smaller gamut, and Canon used a more efficient distribution of code values that accounts quite well for the dynamic range of the C100 and C300.

Here's my analysis of V-Log L banding: http://personal-view.com/talks/discussion/comment/206020#Comment_206020 But it's exactly the same story for little pink and cyan blotches: not enough precision for the large dynamic range and gamut of V-Log, most noticeable in the red color difference channel.

 

[David Del Real]

...I would invite someone to communicate with the beta testers who've been using VLOG-L on the GH4 for the better part of a year; it would seem like those with the most experience would have run into all these problems and already have worked out solutions to them.

Yes, brilliant.

 

[Barry_Green]

It's not just an 8-bit problem. It's the problem of 8 bits being used in a color space with a very wide dynamic range and a large color gamut. V-Log was designed for 10 bits from day 1, and it doesn't make particularly efficient use of code values. Canon Log was optimized for 8 bits. It has a smaller dynamic range and smaller gamut, and Canon used a more efficient distribution of code values that accounts quite well for the dynamic range of the C100 and C300.

Here's my analysis of V-Log L banding: http://personal-view.com/talks/discussion/comment/206020#Comment_206020 But it's exactly the same story for little pink and cyan blotches: not enough precision for the large dynamic range and gamut of V-Log, most noticeable in the red color difference channel.

I'm not disputing that 8-bits (and especially 8-bits when used to encode a wider dynamic range, and most especially 8-bits when only about 150 code values are used) contributes or exacerbates banding issues. That much is obvious, and 10-bit goes a long way towards overcoming that.

What I'm saying is that the purple/green splotches are not an 8-bit problem. They most definitely, unequivocally, are not. And they're not a VLOG-L problem, because the DVX200 uses the same VLOG-L, in the same 8 bits, and doesn't cause any splotches like the GH4 is doing. Whatever the problem is, it is not inherently tied to 8-bits, or VLOG-L, or AVCHD, or even the combination of all three, because another camera can use all those same components and render footage that does not have the problem.

 

[Xrayspecs]

If you blew $99 on the V-Log L firmware update, like I did, I encourage you to complain loudly to customerservice@us.panasonic.com. I did, and got this reply:

Hello,

Thank you for your inquiry.

My sincerest apologies that your experience with Panasonic was less than expected. This is not how we want Panasonic reflected, your complaint has been noted and we appreciate you bringing this to our attention. Please contact customer support at 1-800-211-7262 for updated firmware.
Thank you
Panasonic Direct Sales Support
Call Toll Free: 800-405-0652
Monday - Friday, 8:00 am to 8:00 pm, EST
Saturday-Sunday, 12:00 pm to 5:00 pm, EST

Yeah, I realize it's just a form reply, but if they get enough of these customer complaints, maybe they'll be more motivated to fix the issue. The squeakier our wheels, the better...

 

[refocusedmedia]

I recorded a video of my son at hockey practice yesterday, in FHD on my LG G4 phone. When I uploaded it to FB and play it back, the ice is full of magenta/green blocking. Thanks a lot, you guys.

 

[CustomerGH4]

I recorded a video of my son at hockey practice yesterday, in FHD on my LG G4 phone. When I uploaded it to FB and play it back, the ice is full of magenta/green blocking. Thanks a lot, you guys.

That is nothing, when I go to sleep and close my eyes I see green and purple spots everywhere.
Thanks a lot guys!

 

[Lpowell]

With Canon C-Log, highlight saturation is at 109 IRE and with Panasonic V-Log-L, highlight saturation is at 79 IRE and we can all see what a difference that makes.

With all due respect -- I don't think you can. The difference between the 79 and 109 IRE cutoffs does not explain the purple/green in the lower 9 stops of the gamma curve, where both CLOG and VLOG-L encode fundamentally identically.

Absolutely regardless of Canon's marketing, the fundamental fact is that both CLOG and VLOG-L record 12 stops. They record the first 8 stops fundamentally identically. They record the last 4 stops differently, with the Canon using more codes. Do you disagree with any of this? I would be shocked if you do, but I do have to ask to see if we're at least on the same page to this point.

Let's examine those tone curves more critcally. IMO, it's a stretch to claim those 12-stops are all that Panasonic and Canon's marketing imply:

First, let's identify the part of the curve that is actually logarithmic. It's not what you refer to as "the first 8 stops", that's the non-logarithmic shadow roll-off region on the bottom left of each graph. The logarithmic section of the curve is the flat diagonal line that extends downward from the top right corner of each graph. For Canon C-Log, it starts at about stop +5.5 and remains flat down to about stop -0.5, a 6-stop range. For Panasonic V-Log-L, that 6-stop log range starts at stop +4 and extends down to stop -2. Notice how there's a 1.5-stop difference between what Canon labels as "stop 0" and Panasonic's version? That's because "stop 0" is a meaningless reference point that camera manufacturers shift around to suit their marketing claims.

Regardless of what you call "stop 0", those top 6-stops contain all highlights and most of the midrange. In Canon 8-bit C-Log that range is encoded in over 180 levels, about 30 steps per stop. In Panasonic 8-bit V-Log-L, there's only about 120 levels, just 20 steps per stop. C-Log thus has 50% more numerical precision than V-Log-L in the logarithmic range of its tone curve, and I think that makes a significant difference in image quality.

At the bottom left of each curve is the shadow region, where the dynamic range is claimed to be extended down to 12 stops below the highlight clipping point. But take a hard look at the bottom two stops (Canon: stops -6.5 to -4.5; Panasonic: stops -8 to -6). Those curve segments are nearly horizontal, nowhere close to logarithmic, and are allotted only three 8-bit steps per stop. That is such a crude level of shadow digitization that virtually no detail will be perceptible in those bottom two stops. While a 10-bit or better encoder might preserve something useful, with 8-bit encoding, the bottom two shadow stops might as well be black.

 

[Brandon Freeman]

Gotta say... I'm not backing down on wanting to figure out/resolve the magenta issues, but I'm getting the hang of this profile more and more every time. Either my eyes are adjusting and filtering out the artifact, or as I learn to expose/grade it in its fragile 8-bit state, I'm getting more and more familiar with what it needs to give me the results I want.

Having said that... I still really want a 10-bit 4K recorder!

 

[balazer]

What I'm saying is that the purple/green splotches are not an 8-bit problem. They most definitely, unequivocally, are not. And they're not a VLOG-L problem, because the DVX200 uses the same VLOG-L, in the same 8 bits, and doesn't cause any splotches like the GH4 is doing.

Here is Panasonic GH4 V-Log L footage, recorded internally. Note the greenish splotches in the middle of the bear and on its head. (click for a full-size view)


I've decoded the blue color difference channel, measured its color values in those areas, and increased the contrast (I was aiming for the same frame, but I'm not quite sure if I got it. It's close, anyway):


and the red color difference channel:


Most of that bear is white, but there are colored splotches in which the red and blue color difference channels are just one color value away from the surrounding area. Where there's less blue color difference, it looks yellow. Where there's more blue color difference, it looks purple. Where there's less red color difference, it looks green. Where there's less red and more blue color difference it looks blue.

With 8-bit precision there are just no shades between the green, blue, yellow, and purple of those splotches, and the white of the surrounding area. Those colors look noticeably different to my eyes. But 8 bits just doesn't have the precision to show any finer gradation than that in V-Log. Noise and compression noise make this problem worse, but it wouldn't go away even if you had a perfect exposure and uncompressed 8-bit recording.

Any natural image will have parts that are shifted towards one hue or another, but very subtly and almost imperceptibly. The camera's sensor captures that. But without adequate precision in the recording, even some small difference in the color is going to cause the color value to cross the threshold to a neighboring color value. That's not the camera's fault. The camera must always choose the closest available color value. The problem is that neighboring color values are quite noticeably different colors after decoding from V-Log.

If you're still not convinced I can perform the same analysis for other samples. I haven't looked at any DVX200 footage, so I won't speculate about why you aren't seeing any problems there.

 

[CustomerGH4]

Quick fix, if you have kept the image flat I would have gotten it even better since you can still see it but at least it does not jump into your face.

http://www.upload.ee/image/5215567/supaman.png

 

[ozmorphasis]

Here is Panasonic GH4 V-Log L footage, recorded internally. Note the greenish splotches in the middle of the bear and on its head. (click for a full-size view)


I've decoded the blue color difference channel, measured its color values in those areas, and increased the contrast (I was aiming for the same frame, but I'm not quite sure if I got it. It's close, anyway):


and the red color difference channel:


Most of that bear is white, but there are colored splotches in which the red and blue color difference channels are just one color value away from the surrounding area. Where there's less blue color difference, it looks yellow. Where there's more blue color difference, it looks purple. Where there's less red color difference, it looks green. Where there's less red and more blue color difference it looks blue.

With 8-bit precision there are just no shades between the green, blue, yellow, and purple of those splotches, and the white of the surrounding area. Those colors look noticeably different to my eyes. But 8 bits just doesn't have the precision to show any finer gradation than that in V-Log. Noise and compression noise make this problem worse, but it wouldn't go away even if you had a perfect exposure and uncompressed 8-bit recording.

Any natural image will have parts that are shifted towards one hue or another, but very subtly and almost imperceptibly. The camera's sensor captures that. But without adequate precision in the recording, even some small difference in the color is going to cause the color value to cross the threshold to a neighboring color value. That's not the camera's fault. The camera must always choose the closest available color value. The problem is that neighboring color values are quite noticeably different colors after decoding from V-Log.

If you're still not convinced I can perform the same analysis for other samples. I haven't looked at any DVX200 footage, so I won't speculate about why you aren't seeing any problems there.

Is it possible that the green splotches are reflections from Superman on the highly reflective plastic wrap?

 

[ozmorphasis]

Here is Panasonic GH4 V-Log L footage, recorded internally. Note the greenish splotches in the middle of the bear and on its head. (click for a full-size view)


I've decoded the blue color difference channel, measured its color values in those areas, and increased the contrast (I was aiming for the same frame, but I'm not quite sure if I got it. It's close, anyway):


and the red color difference channel:


Most of that bear is white, but there are colored splotches in which the red and blue color difference channels are just one color value away from the surrounding area. Where there's less blue color difference, it looks yellow. Where there's more blue color difference, it looks purple. Where there's less red color difference, it looks green. Where there's less red and more blue color difference it looks blue.

With 8-bit precision there are just no shades between the green, blue, yellow, and purple of those splotches, and the white of the surrounding area. Those colors look noticeably different to my eyes. But 8 bits just doesn't have the precision to show any finer gradation than that in V-Log. Noise and compression noise make this problem worse, but it wouldn't go away even if you had a perfect exposure and uncompressed 8-bit recording.

Any natural image will have parts that are shifted towards one hue or another, but very subtly and almost imperceptibly. The camera's sensor captures that. But without adequate precision in the recording, even some small difference in the color is going to cause the color value to cross the threshold to a neighboring color value. That's not the camera's fault. The camera must always choose the closest available color value. The problem is that neighboring color values are quite noticeably different colors after decoding from V-Log.

If you're still not convinced I can perform the same analysis for other samples. I haven't looked at any DVX200 footage, so I won't speculate about why you aren't seeing any problems there.

Is it possible that the green splotches are reflections from Superman on the highly reflective plastic wrap?

Edit: Nevermind. I got a better look on my big monitor.

 

[wwjd]

I got this:


Actually, that has nothing to do with VLOG, but was an old test of SLAMMING a really, really dark image to see what happens. Hey! Purple and green. Might tell us something. This is UHD internal 30p 4K


the original:

 

[Frantikthegod]

Just for kicks. Screen caps for an upcoming music I shot using only V-Log L on the GH4 recording externally to an Atomos Ninja Assassin. Metabones Speedbooster XL, Sigma 18-35mm f1.8. Not sure of your preferred way to export screen grabs though.

 

[Murph100]

Gotta say... I'm not backing down on wanting to figure out/resolve the magenta issues, but I'm getting the hang of this profile more and more every time. Either my eyes are adjusting and filtering out the artifact, or as I learn to expose/grade it in its fragile 8-bit state, I'm getting more and more familiar with what it needs to give me the results I want.

Having said that... I still really want a 10-bit 4K recorder!

Care to share how you're grading ? Luts etc

 

[wschmid]

x

 

[dustylense]

You guys should check out Luke Neumann's VLOG Luts. I'm playing around with them and his ETTR LUT is doing wonders.

http://neumannfilms.net/?product=gh4-vlog-l-luts

 

[Lpowell]

I looked through my archive of GH2 Flow Motion v2.02 test footage for scenes likely to provoke the kind of chroma smearing seen in GH4 V-Log-L footage. This slider shot features a dingy, low-contrast background (behind the rusted metal grate) of a type that often produced mottled magenta/green chroma patterns on the unhacked GH2. This footage was shot in 1080p24 at 100Mbps GOP-3, with near-optimal I-frame bitrates, representative of the finest image quality I was able to achieve:

The original GH2 video file can be downloaded from the link. Here's a frame grab from the video: