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
  • @alcomposer

    Panasonic soon will be business of selling bare cameras for $100 and later rent you packages of functions for $20 per month.

  • @Vitaliy_Kiselev :-) That seriously is the most AMAZING comment I have ever read on PV! Called: Camera-Cloud.

    Mind you- if they forget about rental and charge $99 outright, it may work.

    It is sad now that suddenly improving a cameras firmware is not 'breach' of contract but they will probably call it 'stealing'. Panasonic are aware that they are in competition with other camera companies correct?

  • @Vitaliy_Kiselev

    As conversion function exist (LUT is just another form of function) and it will work with ANY (underexposed or overexposed) source data the same. As I said - if you underexposed footage you will get underexposed result, it is all same.

    The knee in the LUT gives it a kink that makes it a non-linear curve. For that reason, it will mathematically match the footage precisely at only one specific exposure level. Of course, you could correct an exposure mismatch by adjusting the gain of the footage before feeding it into the LUT, but that would require manual calibration of your post production levels. The LOG-LUT procedure is intended to be a turnkey workflow that requires no technical knowledge on the part of the user.

  • @LPowell and Vitaliy_Kiselev, I think you guys agree but don't know it :-)

  • @Brumbazz I think there needs to be a dedicated LOG thread- this has become epic. In a good way-September 2015 will always be known as the great LOG debate.

  • The knee in the LUT gives it a kink that makes it a non-linear curve. For that reason, it will mathematically match the footage precisely at only one specific exposure level.

    I think that you are confused here. Function is non linear for any given interval (it looks linear due to representation of x axis on your chart :-) ). And it does not matter at all.

    Again, functions (like one in Panasonic PDF) work for any footage, underexposed and such. Result will also look underexposed (or overexposed in another case).

  • You guys need to say which kind of LUT you are talking about.

    A camera-to-display LUT is designed for footage shot with a certain exposure level. There are push and pull LUTs designed to properly transform footage shot with a lower or higher exposure level compared to normal.

    A camera-to-linear (or camera to true log) LUT doesn't care how the footage was exposed. There's just one LUT for all exposure levels. Exposure compensation is easily applied in a linear or log color space.

  • @Vitaliy_Kiselev It's linear in the semi-log coordinate space of the LUT, which means the straight-line portion of the LUT will scale proportionately with exposure level, as you say. It's the non-logarithmic shadow area that won't scale properly, not just the curve of the knee, but the pedestal level will be off as well.

    @balazer We're talking about Panasonic's V-Log-L, which requires a camera-to-display LUT calibrated to a specific exposure level.

  • It's linear in the semi-log coordinate space of the LUT, which means the straight-line portion of the LUT will scale proportionately with exposure level, as you say. It's the shadow area that won't scale properly, not just the curve of the knee, but the pedestal level as well.

    I do not care about shadow or not, I am talking about math.

    Let's move all to dedicated topic, here it will be only test and firmware parts.

  • Really i dont know how work the LUT.

    Hear an interesting use of Lumetri, not is with V-LOG, but can be interesting for general use in post.

  • Now Im seeing the video "Filippa dancing (GH4R V-Log L footage)" and I think the V-LOG make GH4 can compete with A7S in video (out the astronomic ISO). In the A7S the S-LOG work to ISO 3,200, in the GH4 to what ISO amount?

  • @lpowell @Vitaliy_kiselev is not disputing that there is a complex transformation. He stated there is a known transform f(x) and hence there is an inverse transform g(x)

    Sensor->f(x)->mp4

    We take this mp4 file and

    Mp4->g(x)->prores, CineForm...

    So what do we gain from log d,c,v whatever... Mp4 is 8bit so we are simply destroying shadow data and giving more bits to mids and highlights.

    There is no gain!!

    10bit may be a different story. The 99$ is best saved and invested in an external recorder

  • For me I also think I skip the V-log. It may be very beneficial for some real professional fimemmakers, but not for me.

  • I might just stick to Natural or Portrait. I won't be able to afford a 10-bit Pro Res/ DNxHR Shogun or Ninja assassin to use a LUT to monitor exposure with. Also, I'm not hearing a lot of good conjecture about V-Log L on the 8-bit internal 4:2:0 codec.

  • B&H now have the expected release date of the V-Log L Activation Code as October 7th

  • @zcream

    So what do we gain from log d,c,v whatever... Mp4 is 8bit so we are simply destroying shadow data and giving more bits to mids and highlights.

    Yep, in the V-Log-L thread I pointed out that when converted into 8-bit H.264 format, the bottom four stops of log profile dynamic range are encoded with just 4 bits to cover all four stops. Such a crude level of digitization makes the effort to capture 12 stops of dynamic range futile.

  • Such a crude level of digitization makes the effort to capture 12 stops of dynamic range futile.

    I do not know that it means, really :-). As you can capture in 8bit (and even lower) 20 stops or 40 stops, or 100 stops. They will have tonal issues, but DR won't go anywhere.

  • @Vitaliy_Kiselev

    Yes, that's exactly what I mean, you can pack a wider dynamic range into an 8-bit data space, but only by sacrificing the tonal range of some segment of that dynamic range. In the case of V-Log-L, it is the tonal range (i.e. image quality) of the bottom 4 stops of shadow dynamic range that is sacrificed.

  • In the case of V-Log-L, it is the tonal range (i.e. image quality) of the bottom 4 stops of shadow dynamic range that is sacrificed.

    Not only, log portion also sacrifice data :-) Just math.

  • @lpowell, you've got it backwards. V-Log doesn't capture an extended shadow range. It captures an extended highlight range. Preserving shadows is not futile. It's doing exactly what it should be doing, devoting a small number of bits in the deep shadow part of the V-Log curve for what will be a small number of bits in the corresponding range of the display-referenced final product. The deep shadows are noisy, not because of limited precision, but because of shot noise and sensor noise. You don't want to devote a lot of bits to encoding noise, and can't put your midtones there. That part of the range must be reserved for the deep shadows, and you must expose with that in mind. The bottom two stops will be mapped very close to 0% in the Rec.709 output. Remember that Rec.709 goes right to black, to linear 0.

    In practice the limited precision of the V-Log curve will be a non-issue. You'll be more limited by the camera's noise performance.

    It's true that V-Log is not optimized for 8-bit recording. But it's not because of the shadows. Every camera log space does the shadows in a similar manner, being close to a power function. V-Log is not optimized for 8-bit recording because of the large log section. Log encoding is not perceptually uniform, and so it's inefficient. Some other camera log spaces were designed for 8-bit recording from the start, and they do the more efficient thing which is to keep a power function for a greater part of the recorded range before transitioning to something closer to log.

  • @balazer It extends both, but not so much on the highlight side, and a bunch on the shadow side. look at the graph. Unfortunately the huge highlight extension is only accessible by the Varicam and it's 16 stop DR. The GH4, doesn't get but only a stop on top. 4 total on top, and 8 beneath.

    but that being said, the highlights are very recoverable on the footage I've been playing with... like the banister on that beach shot, you can power window that and get every bit of detail back.

    http://pro-av.panasonic.net/en/varicam/common/pdf/VARICAM_V-Log_V-Gamut.pdf

  • @balazer

    So long as you don't overexpose your footage, there's no such thing as an "extended highlight range". All cameras will capture the entire highlight range up to the point of sensor saturation, at the finest digitization quality the camera is capable of. What happens after that is determined by the camera's tone curve. In the case of V-Log-L, the linear response of the image sensor is converted into a logarithmic curve, preserving midrange and highlight details in a manner similar to the response of the human eye. Ideally, this logarithmic response would be maintained throughout the entire 12-stop range, but technical limitations make this impractical. That is why a non-logarithmic rolloff knee is imposed on the bottom 4 stops of the shadow range, and the entire curve is offset with a 12.5% pedestal. This 12-stop DR spec thus defines how deeply the dynamic range extends into the shadows before being clipped off at black.

    Your claim that "you don't want to devote a lot of bits to encoding noise" is absolutely the wrong approach to preserving shadow detail. The relative increase in noise at low light levels makes shadow details appear more complex to the encoder, and that inherently requires more bits to perserve with high fidelity. In the case of V-Log-L, an 8-bit H.264 encoder will crush this complexity into 4 bits of crude, blocky patterns that obliterate the bottom 4 stops of the dynamic range. What you want instead is an encoder with 10-bit or better precision, and enough bitrate to preserve as many fine details as practical, including as much of the low-level noise as you can preserve. You may then use sophisticated noise management tools in post to extract the highest quality images your workflow can produce. H.264 encoders generally do a lousy job of this, because they are consumer products tuned to favor highlight details and crush shadows into high-contrast blackness. While that may look sharp to untrained eyes, it falls apart when you attempt to lift those crudely digitized shadows up out of the muck. If you don't take care to preserve the tonal depth of your shadows, that vaunted 12-stop dynamic range will be all for naught.

  • @shian, the shadow ranges of Rec.709 and of V-Log go all the way to black. (linear zero) Their shadow ranges are unlimited, effectively. There's nothing to extend. There's nothing blacker than black. That log plot could just as well start at 20 or 40 or 1000 stops below middle grey, an the V-Log curve would extend all the way to the left edge.

    The range of a curve is distinct from the range of a camera. Incidentally, the Varicam 35 doesn't use the entire V-Log range. Its highlights stop at about 6.5 stops above middle grey.

  • @balazer

    the shadow ranges of Rec.709 and of V-Log go all the way to black. (linear zero) Their shadow ranges are unlimited, effectively. There's nothing to extend.

    A totally misleading claim. The dynamic range of V-Log-L extends downward to 12-stops below the highlight clipping point of the image sensor. All DR below 12-stops is crushed to black. That crushed range is not "unlimited", it is obliterated.