Brandon Freeman
Well-known member
True that! Man, if that was the default look out of the camera none of this would matter.
GH4 Exposing V-LOG L
- Thread starter Bern Caughey
- Start date
True that! Man, if that was the default look out of the camera none of this would matter.
Barry_Green
Moderator
It certainly seems to be a GH4-specific issue. Sony's FS7 uses XAVC-L with 4:2:0 8-bit, and log recording, and doesn't show the issue. The other cameras mentioned seem to get fine with 8-bit, so it doesn't seem to be 8-bit's fault.
Reading through the other threads, it seems like everyone generally agrees that the purple/green is always there in the GH4, it just doesn't manifest itself clearly because in other profiles nobody ever pushes the footage as hard as they do in VLOG-L, right? And inherently with VLOG-L, it records very low color saturation, which necessitates pushing the color further to restore proper saturation, and it seems like post-pushing the color is what's exaggerating the purple-green into visibility and objectionable-ness, yes?
If so, I don't know exactly how this is going to be sorted, but I can say this -- the DVX200's original VLOG-L profile was very flat indeed; they actually modified it in one of the most recent firmware updates to record richer color saturation. The result is that you don't have to push it as far in post. Maybe the GH4 will inherit a similar change?
Question, as I don't know the answer -- has anyone here actually received and used the legitimate, paid-for version of VLOG-L? Or is everyone using the app-based back door version? I guess it's possible that the paid-for version would be the new profile as implemented on the DVX200 that involves richer color saturation, which might minimize or eliminate the problems that people have found maybe?
Maybe, maybe not. Again, AFAIK the other cameras record YUV and they don't have a problem. The DVX200 uses (presumably) the same basic encoder, and doesn't exhibit the problem. I doubt that the blame can be laid on 8-bit, or on the bit depth of the VLOG-L profile, or on YUV encoding. If any of those were true (and especially if all of those were true) then we should be seeing the same issue on other cameras and especially on the DVX200. But we don't -- therefore, I say that the evidence is pretty compelling that those aren't the problem. If I had to place a bet on what the problem is, I'd say that it's the level of color saturation adjustment necessary to bring the colors back up to suitable levels. Others have said that the magenta/green exists in all GH4 profiles, but -- nobody ever noticed, and the GH4 has been roundly praised for how good its footage looks, right?
It's possible that a similar color-boosted VLOG-L profile might find its way to the GH4 in the paid firmware version (or in an update thereafter) and that might solve all the problems. Boosting color internally will always yield superior results to trying to boost it after it's been 4:2:0-subsampled, 8-bit quantized, and AVCHD-encoded.
As of the current firmware, the histogram clips at 80%, and the zebras will not show up past 79 IRE, when using VLOG-L. When using other gammas they have the full range of 0-109.
Last edited:
visceralpsyche
Well-known member
Thanks for all the info and thoughts Barry - it's especially interesting to hear about the newer V-Log in the DVX200 that adds saturation back in.
From what I'm hearing, paid for, official firmware versions are identical to the back door version (I'm using the back door version for my testing).
Hopefully Panasonic is reading all this and figures out the solution!
Weirdly enough, on the GH4 the histogram is going from 10 IRE to 100 IRE, but Zebras clip at a maximum of 79 IRE.
Cheers,
Paul
From what I'm hearing, paid for, official firmware versions are identical to the back door version (I'm using the back door version for my testing).
Hopefully Panasonic is reading all this and figures out the solution!
Weirdly enough, on the GH4 the histogram is going from 10 IRE to 100 IRE, but Zebras clip at a maximum of 79 IRE.
Cheers,
Paul
MichaelTiemann
Active member
My GH4 is running authorized Vlog-L on version 2.4 firmware.
morgan_moore
Major Contributor
Ive only shot 10bit FS7.. but go thin with your 'negative' and it does get pretty gack down the bottom end.
Brandon Freeman
Well-known member
My version is running authorized V-LOG L 2.4 Firmware as of Saturday. Have only done one shoot with it (just this morning), and it's still there.
It's all of the above, converging to reveal how a consumer-grade H.264 encoder falters when pushed past its limitations. The reason it's worse on the GH4 is because Panasonic limits V-Log-L to 79 IRE. The reason you see it on smooth gradients is because H.264 shortchanges low-contrast areas in favor of sharp details. The reason you see chroma smearing is because H.264 UV channels are bit-starved by design. The reason you see magenta/green hues is because those are "primary" colors in the YUV color space. The reason you need to boost color saturation in post is because it was suppressed in order to maximize dynamic range. None of these things are "errors" and if you expose and grade your footage carefully, no one will notice. But when they all come together at once...
You left out the cost of media, batteries, cables, support structures, etc. which when all added up will total thousands of $$$
Here's a convenient kit for the PIX-E5H that costs $1100.
http://www.bhphotovideo.com/c/produ...x_e5_e5h_kit_portable_video_recorder_kit.html
Here's the cheapest option for recording media at $250 but I'll probably need at least two so that's $500
http://www.bhphotovideo.com/c/produ...ve_media_enclosure_with.html/pageID/accessory
So now I'm looking at spending $2,795 in order to use the V-log which is supposed to work with internal recording.
Barry_Green
Moderator
And yet... all of those same circumstances exist in the DVX200, and the problem simply doesn't happen there. VLOG-L is limited to the same 79 IRE; and even if it wasn't, the range from 80+ only affects the top two stops and has nothing to do with the first 12. h.264 uses the same processing on both cameras (at least when using AVCHD). The h.264 UV channels can be presumed to be processed the same on both cameras, you would presume. Yet -- the problem doesn't happen on the DVX200.
Which means that none of these things can be pointed to as being the definite source of the problem, and neither can they be used to say that the GH4 is inherently faulty. If there is a problem, it must be attributable to something else. Is there some deficiency in the GH4 that keeps it from being able to perform at the level of the DVX200? Perhaps. Do the beta-testers who have been using VLOG-L on the GH4 for nearly a year acknowledge this as an issue? If not, then -- what are they doing different?
wschmid
Well-known member
I think that is a nice test - but for sure the GH4 will not be able to compete with the Red! Would be a miracle really!
Sure. But the 8bit limitation derives by the limitation of v-log v with IRE79. And I think there is no way to overcome that by adding something on top of the limited v-log curve. The only solution would be another log curve, using 0..255 in 8bit similar to s-log curves. Since that will not happen I think the next best solution is to record external with 10bit 422 and use this footage to grade it as one will need it.
Barry_Green
Moderator
Not true though. Other log curves use the same (or extremely similar) encoding between the darkest stop and "90% white". The only limitation VLOG-L has is in recording the stops up above 90% white; VLOG uses the range from 61 to 79 IRE for all that, and other curves can use the range from 61 to 109 IRE. So the only place you'll see any difference between VLOG-L and SLog or CLog or (probably) J-Log is going to be in the two or three brightest stops. All the rest is being recorded fundamentally the same in all the various log gamma curves.
SLOG does not use 0-255. SLOG uses 23-255. Log curves generally don't use the range from 0-23, since it would be adversely affected by the video compression algorithms used (specifically, MPEG-4 h.264).
Wrong conclusion. It's not only the brightest few stops that are shortchanged to near 7-bit precision in 8-bit V-Log-L. The entire 12-stop range is scaled down to fit within the truncated 32-180 scale, degrading the tonal resolution of every stop along the way. The GH4's internal H.264 encoder is not well-optimized for deliberately underexposed footage, regardless of whether it's V-Log or a standard profile. It's a consumer-grade encoder tailored to produce brightly-colored, high-contrast home movies for direct viewing on laptops and HDTV's. It's frankly remarkable it works as well as it does, considerating how severely mismatched V-Log-L is to Rec 709.
visceralpsyche
Well-known member
Exactly, that's what bothers me most about the profile not using the entire 0-255 range. It's a storage range, so compressing the data much more than the linear profiles (which do use the 0-255 range) seems to be making for a much worse picture to our eyes, even if the dynamic range is technically higher.
It's not a tradeoff I'm willing to make for internal recording, so for me personally V-Log L is useless unless I buy a 10 bit external recorder (which I may yet do, but for reasons beyond simply getting V-Log to behave).
Cheers,
Paul
$1,195 Video Devices PIX-E5H
$ 59 Video Devices SpeedDrive Media Enclosure ( Empty )
$ 192 Samsung 500GB 850 Evo mSATA SSD
$ 120 Two : Watson NP-F975 Lithium-Ion Battery Packs
$ 40 Two : Watson battery chargers
Total Cost = $1,606
For 500 GB of storage space that will likely run all day on two F975 batteries.
Since I already own the batteries and chargers that I use for other gear, my cost will be $1,446 + tax total.
Barry_Green
Moderator
This is simply not true.
The scale is exactly (or near unto exactly) the same for VLOG-L as it is for CLOG, and the various versions of SLOG. The code values for 90% white and 0% black are the same between them all. If you take a gamma like CLOG, which delivers 12 stops, and expose it the same as the VLOG-L on the GH4, you'll get exactly (or nearly exactly) the same code values from pitch black to 90% white. The only area you'll see any difference at all is in the stops above 90% white, where the VLOG-L will compress them into 62-79 IRE, and CLOG/SLOG etc will have room to spread them out in the range of 62-109 IRE.
Barry_Green
Moderator
No log gamma uses the entire 0-255 range, that I know of. They all have a floor of around 32, although I think SLOG1 goes down to 24. They all know they're going to be compressed by video compression, so they set a pedestal below which they do not use that range, to avoid what video compression does to that lower range.
Sorry, but I don't think Canon's marketing brochures can persuade the GH4's V-Log-L implementation to work any differently in practice. Let's compare Canon's C-Log tone curve to your diagram of Panasonic's V-Log-L:
Canon claims its C-Log "90% reflectance" point is at 62 IRE, but that puts full-scale 109 IRE at the absurdly inflated level of "720%" reflectance. That's a completely arbitrary number that Canon selected to make it sound like its cameras have a humongous DR beyond 100% reflectance, which happens to be physically impossible, since reflectance is defined on a 0-100% scale:
https://en.wikipedia.org/wiki/Reflectance#Reflectivity
In real-world terms, every image sensor has a white saturation point where highlights max out, and the crucial engineering decision is how to map this point on the IRE tone curve. In the C-Log and V-Log tone curves above, neither curve uses a knee to roll off the highlights, they both remain flat and logarithmic up to the max F-stop. That means that both curves' highlight exposure will scale proportionally. With C-Log, Canon scales the sensor's white saturation point to nearly 109 IRE. But with V-Log-L, Panasonic scaled the GH4's white saturation point down to 79 IRE, as you indicated with the green lines on the V-Log chart. Compared to the 109 IRE scaling used by Canon, Panasonic's 79 IRE scaling degrades the tonal precision of V-Log-L recordings, particularly when using an 8-bit H.264 encoder.
Barry_Green
Moderator
The curve is not the exact same shape, but the relevant points are. If you read the Canon white paper on their C-LOG curve, it states in there that the middle gray point is adjustable based on exposure, depending on how many stops you want to calculate above and below middle gray.
If you look at both charts and pay attention only to the code values, CLOG starts at 128 (in 8-bit, that's 32). Canon's CLOG starts at 7 stops down from middle gray, Panasonic's starts at 8 stops down. However, if you look at the code values, middle gray on the Panasonic is 448 (using 10-bit codes to make it easier), and Canon's curve intersects 1 stop up from middle gray at ... wait for it ... 448. Which means that the first 8 stops (-8 to 0 on the VLOG, -7 to +1 on CLOG) are encoded in exactly, exactly the same code values, 128 to 448. There is NO penalty for the first 8 stops, at all, due to VLOG's early cutoff point.
Both curves go on to encode four more stops above that point, but from the point of code 448 on down, they are both encoding 8 stops in exactly the same range.
Canon encodes 5.3 stops above middle gray (but, again, their version of middle gray is 1 stop below where Panasonic puts it). Panasonic encodes 4 stops above middle gray. Canon hits 90% reflectance at code value 614. Panasonic hits 90% reflectance at 602. Almost identical code values. Now, Canon allocates 2.25 stops above middle gray to 90% white. Panasonic allocates 2.25 stops above middle gray to 90% white. So Canon has now encoded 9.25 stops in the range of 128-614; Panasonic has encoded 10.25 stops in the range of 128-602. Is there a difference? Yes, a little bit of difference -- but only in that last stop. We've already shown how the first 8 are identical. So there are fewer code values in the brightest stops, yes. It is as crippled as you seem to want to describe it? I really, really don't think so. The first 8 stops are encoded with an identical (yes, identical) amount of code values between the two. So the only place where there's any difference at all is in the top 4 stops. Panasonic encodes the top 4 stops in a range from 448 to 730-ish. Canon encodes their top 4 stops in a range from 448 to 1016. Yes the VLOG-L cutoff limits the amount of code values for the last four stops, but only for those last four stops. The rest of the range is encoded with EXACTLY the same precision between the two LOG gammas. And, as I'm sure you'll agree, bits allocated to the brightest stops are the least important, since we visually discern less detail in the brightest stops. What's important is the allocation of shades and bits in the lower stops. And in those cases, both C-LOG and VLOG-L are fundamentally identical in terms of allocatable bits.
As referencing the original VLOG, VLOG-L encodes the first 12 stops exactly the same as the Varicam's VLOG does. The only difference whatsoever is in how the last two stops are encoded; in VLOG-L they just don't exist, the space used for encoding them is the range beyond where the 12-stop sensor can reach.
If you look at both charts and pay attention only to the code values, CLOG starts at 128 (in 8-bit, that's 32). Canon's CLOG starts at 7 stops down from middle gray, Panasonic's starts at 8 stops down. However, if you look at the code values, middle gray on the Panasonic is 448 (using 10-bit codes to make it easier), and Canon's curve intersects 1 stop up from middle gray at ... wait for it ... 448. Which means that the first 8 stops (-8 to 0 on the VLOG, -7 to +1 on CLOG) are encoded in exactly, exactly the same code values, 128 to 448. There is NO penalty for the first 8 stops, at all, due to VLOG's early cutoff point.
Both curves go on to encode four more stops above that point, but from the point of code 448 on down, they are both encoding 8 stops in exactly the same range.
Canon encodes 5.3 stops above middle gray (but, again, their version of middle gray is 1 stop below where Panasonic puts it). Panasonic encodes 4 stops above middle gray. Canon hits 90% reflectance at code value 614. Panasonic hits 90% reflectance at 602. Almost identical code values. Now, Canon allocates 2.25 stops above middle gray to 90% white. Panasonic allocates 2.25 stops above middle gray to 90% white. So Canon has now encoded 9.25 stops in the range of 128-614; Panasonic has encoded 10.25 stops in the range of 128-602. Is there a difference? Yes, a little bit of difference -- but only in that last stop. We've already shown how the first 8 are identical. So there are fewer code values in the brightest stops, yes. It is as crippled as you seem to want to describe it? I really, really don't think so. The first 8 stops are encoded with an identical (yes, identical) amount of code values between the two. So the only place where there's any difference at all is in the top 4 stops. Panasonic encodes the top 4 stops in a range from 448 to 730-ish. Canon encodes their top 4 stops in a range from 448 to 1016. Yes the VLOG-L cutoff limits the amount of code values for the last four stops, but only for those last four stops. The rest of the range is encoded with EXACTLY the same precision between the two LOG gammas. And, as I'm sure you'll agree, bits allocated to the brightest stops are the least important, since we visually discern less detail in the brightest stops. What's important is the allocation of shades and bits in the lower stops. And in those cases, both C-LOG and VLOG-L are fundamentally identical in terms of allocatable bits.
As referencing the original VLOG, VLOG-L encodes the first 12 stops exactly the same as the Varicam's VLOG does. The only difference whatsoever is in how the last two stops are encoded; in VLOG-L they just don't exist, the space used for encoding them is the range beyond where the 12-stop sensor can reach.
Last edited:
So I've read your explanation Barry, and have followed it from start to finish in agreement. But there's just one thing? Where is this horrible macro blocking coming from? Is it the excellent Panasonic codec ruining the party? Has anyone done any test to map out Canon's and Panasonic's true response curves (a lot of work, I know).
As bad as this problem is, has anyone seen a similar issue on Canon C-Log footage?
As bad as this problem is, has anyone seen a similar issue on Canon C-Log footage?
As I explained above, Canon's mathematical manipulations are based on their "720% reflectance" highlight scaling, which is physically impossible. Sorry, but that means your comparison of V-Log to C-Log scaling where you ask us to "pay attention only to the code values" is founded on Canon's misleading marketing bullshit. In reality, Canon C-Log cameras do not have two stops of highlight DR that "just don't exist" on Panasonic V-Log-L cameras; that is nothing more than a number-juggling fabrication. Both Canon and Panasonic cameras record the full highlight range captured by the image sensor, and their log profiles preserve every stop of highlight exposure in logarithmic proportion. Where they differ is purely a matter of how the cameras' tone curves scale the output of the image sensor, starting from max highlight saturation and extending downward toward black. 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.