Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
GH4 Firmware 2.3, V-log for $99, Epic Panasonic marketing fail
  • 1451 Replies sorted by
  • @Vitaliy_Kiselev, I think the thing that is confusing the issue here is that log significantly remaps the actual bit values of the output. In the case of V-Log, the output is clipped at 79IRE, which means that false colour will show 100% clipped pixels as sitting squarely in the upper mids, as will a waveform or histogram display. This cannot be expanded for the purposes of exposure control unless you use a display LUT and meter off the post-LUT values. So, unless you want to record internally at 8-bit AND the GH4 somehow puts a bt.709 LUT on the HDMI output (not sure if it does) you're stuck with the log values coming out the HDMI for preview/metering. To avoid overexposure, you'd need zebras at ~76 IRE, and that's pretty low. False colour would be entirely mis-scaled since most of those expect mids around 50-60 IRE, and the mids in V-Log are about 37-47 IRE, give or take a bit.

    The whole point of LOG is not to put more bits where we think they should be, but to give equal bits per stop of exposure, as befits wide-latitude celluloid film transfer scans. This means, compared to bt. 709 stuff which is what most people are used to, digital log gives actually COMPRESSED midrange in terms of bit-values versus the highlights and shadows. Put another way, since the exposure latitude anticipated by bt. 709 is far less than film you have to put the extra latitude somewhere, and that means taking bits away from the narrower 0-100% reflectance usually catered to by bt.709 and giving them to the extra potential range of the sensor.

    What you end up with is footage where the mids should be sitting at 42 IRE and the highlights WILL clip at 79 IRE, or between 4-5 stops above middle gray. Shadows apparently clip at 7.3 IRE, which is almost 6 linear stops under middle gray, but due to additional compression in the darkest shadows, you are really getting about 7 stops of detail in there.

    If my math is correct, you get approximately 6 IRE per stop of light, which in 8-bit translates to roughly 13 bit-values per stop, or of course four times that in 10-bit.

  • ^^^^^ Thank you

    I felt that clipping was going to be a given, regarding Barry's warning, and the natural instinct to use zebras and WFM, et al, but despite the great extension below, I just know, in an attempt to retain highlights, there's gonna be some peeps capturing some murky shadows.

  • http://www.personal-view.com/talks/discussion/comment/204409#Comment_204409 Last paragraph

    You cannot use "false color" with VLOG-L as it won't show the proper values. Unless you press the "REC 709 LOG" user button, which overlays a REC 709 gamma on the monitor output -- if you do that, then yes you'd see traditional waveform values on your external monitor, and you'd easily see that the image was grossly clipping."

    Comment makes no sense, as false color works in the resulting space exactly as it works always. Yes, if you want to see ranges in final image - you need to make shifts in your head.

    what u see is not even close to what you get. Too many people will not be able to correlate between the camera appearing to be able to see into the shadows just fine with perfect noiseless clarity, and the reality that they are grossly underexposed.

    How about we move into more formal approach?

    We can simplify thing by focusing only on Y component, so usual approach is F(raw)=y1 and log is different function G(raw)=y2. raw value is 12 bit and is almost linear (proportional to number of electrons that hit pixel). Resulting values Y1 and Y2 are 8bit (actually, even less, as range is limited and not [0..255]) and they are not linear.

    @StudioDCCreative

    How about we just get rid of IRE? As this analog video term just complicates things without any necessity.

    It is enough to look at raw sensor values and look at resulting values (after LOG function transform).

    The whole point of LOG is not to put more bits where we think they should be, but to give equal bits per stop of exposure, as befits wide-latitude celluloid film transfer scans. This means, compared to bt. 709 stuff which is what most people are used to, digital log gives actually COMPRESSED midrange in terms of bit-values versus the highlights and shadows.

    Yes, last part is correct. But it also means that ideally we must expose so highlights (in resulting space that you see on monitor will be as close to maximum as possible).

  • @dial That would likely require a bit of processor power. The GH4 is powerful for what it is but I'd be surprised if they could pull it off with just a simple firmware update.
    The only practical solution I can see is if you have a LUT capable external monitor like the Small HD DP7, 502, Odyssey 7, Shogun, or similar monitor, so add about another $1500 to that $99 firmware upgrade if you want to be able to utilize it fully.

  • Can someone explain how similar or different exposing V log on the GH4 will be compared to Blackmagic film. I have both cameras and with my BMPCC I use a Cineroid EVF - internal generic LUT applied, contrast and saturation boosted, zebras at 100%, use false colors for metering skin-tones and consistently get very good exposures with very little noise. From what I have been reading is this method not going to work with the V log on the GH4?

  • But it also means that ideally we must expose so highlights (in resulting space that you see on monitor will be as close to maximum as possible).

    This ends up being not quite true, because there are all different manner of highlights: specular highlights, large flat surfaces reflecting the sun, lamps, bright sky, white objects, etc. Depending on how bright the highlighted objects are and on how bright the rest of the image is, you will want to place each different kind of highlight at a different output level. Some of them should be clipped. A histogram or false color monitor can help, but looking at the levels of highlights alone is one of the more difficult ways to judge the exposure.

    Looking at midtones is better. But the best thing is to really look at the picture as a whole. False color is one tool for judging levels of the entire image, but it requires a clear understanding about how each different part of the camera's log range works.

    The camera's built-in exposure meter is easier to use and more reliable, but less accurate. I see no reason why the GH4's built-in exposure meter couldn't be used. Who knows if Panasonic is mapping middle grey correctly. But you can always compensate.

    In my experience judging exposure on a monitor with properly mapped colors and a normal contrast curve is the most reliable tool for setting the exposure. Judging a camera log image on a normal monitor without any LUT is hopeless. You can be several whole stops away from the ideal exposure, and you'll have no idea. In my opinion a monitor LUT is a necessary tool for getting proper exposure when shooting in a camera log space. And since the GH4 doesn't have one built in and few people will buy an external monitor with one, I'm fully expecting to see a lot of badly exposed and badly graded V-Log footage quite soon.

    Getting the right exposure is really important. A camera log space can represent a large dynamic range, but the GH4's sensor can't capture a large dynamic range, at least not a large usable dynamic range. The shadows are noisy. V-Log L doesn't do a thing to improve the camera's exposure latitude.

    Camera log spaces are good for capture and they can be good for grading, but they're terrible for monitoring. I think Panasonic would have served more users by having something more like Sony's HyperGamma instead. Sony putting S-Log2 on cameras like the a7S and the a7R II without a monitor LUT is just as silly, but at least those cameras also have HyperGamma.

  • This ends up being not quite true, because there are all different manner of highlights: specular highlights, large flat surfaces reflecting the sun, lamps, bright sky, white objects, etc. Depending on how bright the highlighted objects are and on how bright the rest of the image is, you will want to place each different kind of highlight at a different output level. Some of them should be clipped. A histogram or false color monitor can help, but looking at the levels of highlights alone is one of the more difficult ways to judge the exposure.

    Do not complicate things. Meaning here for highlights was as most bright objects you want to not clip in resulting image.

    And I do not see any problems with false color usage. Your head will adjust quite fast.

    Judging a camera log image on a normal monitor without any LUT is hopeless. You can be several whole stops away from the ideal exposure, and you'll have no idea. In my opinion a monitor LUT is a necessary tool for getting proper exposure when shooting in a camera log space. And since the GH4 doesn't have one built in and few people will buy an external monitor with one, I'm fully expecting to see a lot of badly exposed and badly graded V-Log footage quite soon.

    It also make little sense. What is exactly "ideal exposure"? And why you will be "stops away"?

    I do not really like of all this quite simple math things turned into some voodoo (and usage of old analog things like IRE does not help here also).

    Also all this film mentioning is not too relevant, as film sensitivity is based on physics of emulsion reaction to light. And color correction is made on scanned image of the film (that is also digital, same as with digital sensors).

    Here we have almost linear (we can simplify and assume that they are ideally linear) digital sensors. Ideally we want all the information we have. LOG formats are not silver bullet, just one of the methods how to shrink this information into less bits of output.

  • @StudioDCCreative

    If my math is correct, you get approximately 6 IRE per stop of light, which in 8-bit translates to roughly 13 bit-values per stop, or of course four times that in 10-bit.

    Unfortunately, that does not appear to be the case with Panasonic's implementation of 12-stop V-Log-L. On a 10-bit scale, V-Log-L records 0% black reflectance at level 128, midrange 18% gray at level 433, and 90% white at level 602. On a camera with an 8-bit codec like the GH4, you would have a scale from 0-256. That means highlight clipping would be recorded at around 182, with no image details recorded in the range 183-255. At the other end of the scale, the bottom 4 stops of shadow detail would be recorded in the range 32-48, just 4 bits of data to cover 4 stops of dynamic range. Even if we assume, for the sake of argument, that the GH4's H.264 encoder can preserve this data with flawless precision, the decoded result will contain image data limited to a 32-182 range. That leaves 106 unused data levels, nearly an entire MSB of bit-depth, resulting in an effective 7-bits of encoded data.

  • What is exactly "ideal exposure"?

    The ideal exposure is the one that strikes the best balance between preserving the highlights that you want to be not clipped, and minimizing noise.

  • @balazer

    The ideal exposure is the one that strikes the best balance between preserving the highlights that you want to be not clipped, and minimizing noise.

    Indeed, but that is not how V-Log-L is intended to be used (along with a precalibrated LUT in post). In order to match the dynamic range of the LUT, the footage must be exposed with the white clipping point at 79 IRE. That is considerably lower exposure than the GH4's 109 IRE max exposure limit.

  • In order to match the dynamic range of the LUT, the footage must be exposed with the white clipping point at 79 IRE. That is considerably lower exposure than the GH4's 109 IRE max exposure limit.

    Again, any reason to use IRE term at all?

    On a camera with an 8-bit codec like the GH4, you would have a scale from 0-256. That means highlight clipping would be recorded at around 182, with no image details recorded in the range 183-255.

    GH4 does not work with actual 0..255 range in H.264 encoding as far as I know.

    You also again made things so complicated messing reflectance, stops, IRE and values.

    Again, proper way in digital sensor is to view it from the linear raw side and talk HOW this values are transformed. And it is not so hard to make tests even.

    All this reflectance, midranges and IRE is just making mess in the people heads.

  • Indeed, but that is not how V-Log-L is intended to be used (along with a precalibrated LUT in post).

    If you use a "precalibrated" LUT to transform from the camera color space to the display color space, yes, there's only one correct exposure and it's not the one that strikes the best balance between clipping and noise. But I don't use those LUTs. I transform from the camera color space to a linear or log space in which I can do accurate exposure compensation and color balancing. That frees me from the constraints of a camera-to-display LUT.

    ... white clipping point at 79 IRE. That is considerably lower exposure than the GH4's 109 IRE max exposure limit.

    I think you'll find that V-Log L on the GH4 is not limited to 79% RGB output when the ISO setting is raised.

  • @Vitaliy_Kiselev

    GH4 does not work with actual 0..255 range in H.264 encoding as far as I know.

    Actually, the GH4 does provide that option with its H.264 encoder.

    You also again made things so complicated messing reflectance, stops, IRE and values.

    These are all industry-standard terms, used in Panasonic's documentation to define V-Log-L specifications. Log profiles are inherently mathematical constructs and using them as intended requires precise calibration. Otherwise, you're just using it as a subjective flavoring, like any other profile.

  • Actually, the GH4 does provide that option with its H.264 encoder.

    As far as I remember we had post explaining that it just set special flag that tells playback device that it can use fullrange, actual values does not change. In other words, it needs tests (and you need to do it on proper stage, before they will be scaled).

    These are all industry-standard terms, used in Panasonic's documentation to define V-Log-L specifications. Log profiles are inherently mathematical constructs and using them as intended requires precise calibration. Otherwise, you're just using it as a subjective flavoring, like any other profile.

    IRE is fucking mess left from analog days and have no real place in discussion of log profile of digital camera. Yes, some people are very used to it, but for most normal people it just nonsense.

    I also do not understand that "precise calibration" mean. As all I am interested in is to understand how exactly it works (raw to image transform). Panasonic shows some in their PDF, but some tests and such can help anyway.

  • @Vitaliy_Kiselev Absolutely. I use IRE simply to not have to convert back and forth for 8- and 10-bit values. But we can work in bit values too. Really, in the video world (which this all is), it's equivalent. It would be nice to have a more elegant term than "8-bit value" or "10-bit value"... which is why I think we all still use IRE. But I'm sure it can be confusing to some.

    As to your thoughts on false colour, I could agree IF the false colour algorithm in the displays was tuneable, or if the displays were post-log adjusted. The trick is that the reason one shoots log is to be able to make certain adjustments after the fact, so no LUT is going to give an accurate example of what that scene will look like. If the log output via HDMI never exceeds 173 (8-bit value) as that's the white clip point for V-Log L, then the display will never show that as white without a LUT. So unless the display knows that that should be the white clip point (and the meters too) then the user will not realize they are clipping. No false colour that I'm aware of has a clear break at 173 for clip indication, for instance. Same thing for skintones, which usually sit ~55 IRE = 121 8-bit value. In V-Log L, they'll sit around 55-60 8-bit value, looking significantly underexposed to the average user and falling well below the typical false colour skintone range would be. Again, these are all without an interpretive LUT to scale the range back into the BT.709 space those waveform monitors and false colour overlays are calibrated for.

    @LPowell, true, there's a rolloff knee in the last few stops (but really, 8 stops under middle grey is a bit excessive for the noisy GH4 sensor anyway, in my opinion). -6 is quite usable, and if we take the value from -4 (192 in 10-bit values) to the value at +4 (~746 or so, estimated) we get 8 stops in 554 value range = ~69 values per stop. In 8-bit terms thats still just over 17 values per stop. It's the last 4 stops, agreed, that get rolled off pretty hard, but even so, I'd be willing to guess that sensor noise is around par with the quantization noise you'll get down that far anyway - Canon even explicitly chose their log curve to suit the sensor noise - the log curve has less quantization noise than the sensor's ADC, so the limitation is definitely not in the log encoding. Now, I'm not stating this as fact, but from my experience with the sensor so far, it seems a reasonable guess.

    My main concern with V-Log isn't the amount of bits it's giving to the various stops - it's that the clean capture range for the sensor is limited to less than the claimed 12 stops - it's really a 9 stop sensor if we factor in the quantization noise (sensor or log encoding rolloff, take your pick) at the lowest few stops. Now, 9 honest, real, usable stops is nothing to sneeze at for a digital camera, and if the log data isn't falling apart at 8 bits (which I haven't seen evidence of yet in the test footage) then I'd say it's probably well tuned for the sensor. But it does concern me that we're pushing the true limits of the Panasonic sensor technology at the moment.

  • Absolutely. I use IRE simply to not have to convert back and forth for 8- and 10-bit values. But we can work in bit values too. Really, in the video world (which this all is), it's equivalent. It would be nice to have a more elegant term than "8-bit value" or "10-bit value"... which is why I think we all still use IRE. But I'm sure it can be confusing to some.

    Normal terms for digital sensor are - pixel well capacity (in electrons), combined pixel noise (in electrons, temperature and gain dependent), and pixel value (in electrons). I will exclude complications arising from sensor pixels variations. Your potential DR is defined by capacity and noise. And actual is defined by value and noise.

    We can simplify things here assuming that no fixed pattern processing and noise reduction occurs in raw (it is not so, but for simplicity).

    Normal approach is try to preserve linear nature in midtones and add non linear transform to shadows and highlights.

    Log approach is different. But best way for log is to also use full available output range (as doing otherwise is stupid).

    But same tools, as false color perfectly work for it, and you need to just understand how each level range work now.

    P.S. Some people here start to assume that you cam somehow have match between different cameras if you look at raw level and just grade similarly. In digital cameras, in fact, biggest mess happens in the filter above sensor level and also how sensor catch different length light (and it is different, hence Faveon design), so each pixel collects not some perfectly filtered narrow band of spectrum(as firms want you to believe) but has very complicated and wide spectrum filter.

  • @StudioDCCreative

    It's the last 4 stops, agreed, that get rolled off pretty hard, but even so, I'd be willing to guess that sensor noise is around par with the quantization noise you'll get down that far anyway.

    If you discount those bottom 4 shadow detail stops, the vaunted 12-stops of dynamic range offered by V-Log-L is reduced to 8-stops. That's no better than what you get from the standard built-in profiles, making the whole Log-LUT exercise rather futile.

  • @Vitaliy_Kiselev

    Normal terms for digital sensor are - pixel well capacity (in electrons), combined pixel noise (in electrons, temperature and gain dependent), and pixel value (in electrons). I will exclude complications arising from sensor pixels variations. Your potential DR is defined by capacity and noise. And actual is defined by value and noise.

    GH4 does not work with actual 0..255 range in H.264 encoding as far as I know.

    Yes, but we aren't talking about raw sensor data, here. We are talking about video-encoded data. Sure, you know far more about H.264 than I do, internally, but in the end the signal goes out over HDMI to the monitor or recording deck, and that signal, via HDMI, has nothing to do with H.264. It does, however, have everything to do with the 0-255 range you mentioned previously. That range is exactly what is output via HDMI, and that range is the subject of all the questions regarding V-Log L bit-level encoding transforms right now.

    Whether the GH4 running V-Log L scales the upper bound of that 4th stop to something close to the maximum data value at it's given bit depth is an open question, but if it doesn't, then LPowell and Shian are both right in saying that the monitors are going to see only about half of the full range of data values per pixel, and since the metering software in those monitors, including the false colour transform, will be basing their input on values which do not map to the whites and blacks and midtone ranges they are designed for. If it does, then you may be right, and the main difference may be effectively just a strong midtone gamma shift. However, the data coming off the example files right now doesn't suggest that is the case, since their bright whites are seeming to clip at mid-grey.

    So, unless I'm missing something, I'm not sure I see what you're trying to get at -- are you saying that you think the output is scaled to full range?

  • @LPowell that is, precisely, one of my concerns.

  • and that signal, via HDMI, has nothing to do with H.264. It does, however, have everything to do with the 0-255 range you mentioned previously. That range is exactly what is output via HDMI, and that range is the subject of all the questions regarding V-Log L bit-level encoding transforms right now.

    Usually it is not so and data used for HDMI and as buffer intended for compression are exactly same data (for most cameras).

    You can also make tests using HDMI capture to look at the histogram values and if it is limited range or limited range rescaled.

    then LPowell and Shian are both right in saying that the monitors are going to see only about half of the full range of data values per pixel, and since the metering software in those monitors, including the false colour transform, will be basing their input on values which do not map to the whites and blacks and midtone ranges they are designed for. If it does, then you may be right, and the main difference may be effectively just a strong midtone gamma shift. However, the data coming off the example files right now doesn't suggest that is the case, since their bright whites are seeming to clip at mid-grey.

    I think no one is so stupid so if won't learn how false color works now after shot time of shooting log (as it is just different ranges colored differently). It is same as state that ruler does not work now if you measure screw that is two times longer and want to get measurement in US metering values.

    My intention is to simplify this thing, as I see utter pain and so much bullshit across the web concerning this log craze.

  • I'm out. This conversation just keeps going around in circles.

    I guess we just have to wait until VK experiences it for himself. Cuz explaining it, just ain't getting through. log is so completely different from anything you've ever dealt with. It doesn't work the way you think it works.

    Yoda - 'You must unlearn what you have learned"

  • I guess we just have to wait until VK experiences it for himself. Cuz explaining it, just ain't getting through. log is so completely different from anything you've ever dealt with. It doesn't work the way you think it works.

    It is sad, as whole point of conversation was to move from "you need to experience it" to normal formal things.

    Log formal it is simple math that is only slightly different from usual processing that existed in GH4 for years.

    I will be glad to hear how to deal with it at grading stage, but at formal level. Without feelings, emotions or usage of Force.

  • Its interesting how none of the beta testers, such as Nick Driftwood or any others have weighed in on this, if the curve is the Varicam Curve mapped to 12 stops and therefore very easy to get highlight clipping (ac to Barry Green) Why have none of the testers noticed or mentioned this? I am not challenging anyone, I am simply curious why after 9 months of beta versions of VLOG-L on the GH4 given to various people and stores...we only heard about this in the last week or so in relation to the review of the DVX200...just seems a bit strange to me...

  • @shian (apologies if I over-simplify anything, I'm just trying to keep things clear)

    Okay so anything with less contrast (such as any LOG profile) is going to be much harder to judge by eye what is falling where in terms of exposure. You expect that 90% of shooters will have underexposed footage when shooting with V-Log L. So to make it simple, what would be the problem with shooting with the only consideration being given to the highlights. Shoot as close to clipped white as possible, but make sure you don't clip anything you don't intend to. Wouldn't this alleviate the problem you're presenting?

    Obviously metering and monitoring with LUTS is an ideal scenario but on a fundamental level, wouldn't it work to just shoot as bright as possible without clipping white?

    Now I did see this video in regards to the DVX200 and V-Log L which shows clipped white as 80%/bright gray due to the extra 4 stops preserved in regular V-Log (if my understading is correct)

    But even if the GH4 implements a similar V-Log L with clipping at 80%, and I'm only monitoring V-Log L, to my eye I can still tell that that's blown out. If I wanted to preserve those highlights, I could stop down a bit, and make sure I have detail there. As compared to the detail-less gray splotch that's currently there.

    That's a very simplistic way of shooting but if you're not adjusting the lighting, wouldn't that be a sufficient method for achieving a decent exposure?

    And if you're shooting V-Log L, couldn't you switch the GH4 over to Portrait to check your exposure levels and then back to V-Log L for recording? Sort of a poor man's in-camera LUT?

  • The new atomos hdmi recorder has come out. Has 3d lut. Does 422 10bit. Is cheep er than shogun and hdmi only. Buy it an be happy.