precise <> pedantic
With log profiles, you really do need to pay attention to the math. Otherwise, you're just using it as flavoring, like any other profile.
GH4 GH4 V-log?
- Thread starter Kindredmachine
- Start date
joe12south
Veteran
Just a little reminder how different V-LOG L is from Cine-D.joe12south
Veteran
Can it just be here so we can all start tearing apart and fighting about how awesome/terrible our footage looks?
:kali::love4::furious3::evil::thumbsup::2vrolijk_08:rogar-BlackEye(DBG:angry::nads:
Once it's here I'll be too busy shooting and enjoying the results to argue.
Brandon Freeman
Well-known member
Look you either will pay for it or you won't, use it or you won't. Simple...
Hi Illya,
Could you give us any more detail about setting the exposure with V-log, is the histogram mapped to the log profile so clipping is at 100% (79 IRE)? There is some discussion about the need for a monitor with LUT functionality? Thanks
joe12south
Veteran
Whether one uses a LUT or not has no impact on the fact that a V-LOG L encoded image can contain more dynamic range. LUT's are fantastically convenient for getting your grade in the right neighborhood in very short order, but a colorist can also rely on other tools (waveforms, scopes, etc.,) and even their good old eyeballs (with a calibrated monitor) to get to the final image that they desire. It's not even unusual to purposely use the "wrong" LUT to get a desired affect.
Lintel
Active member
That really does look a world better. Both in colour and dynamic range. Lovely. Looking forward to using some Alexa luts!
Polaroid22
Well-known member
But is it recorded 10bit or 8 bit this example, as for run and gun situations I only shoot internal so only 8bit and I dont want to use a recorder?
I would like to see some examples 8bit vs 10bit, cine-V vs V-log.
Too bad there arent a lot of test video's out yet.
I would like to see some examples 8bit vs 10bit, cine-V vs V-log.
Too bad there arent a lot of test video's out yet.
joe12south
Veteran
That's a 10-bit example, but that scene (as with probably the majority of scenes) doesn't offer up a problematic scenario for 8-bit. 10-bits of resolution are always nice to have, but the additional precision doesn't change the fundamental response of the profile.
12-stops of dynamic range sounds great on paper. What's not so great will be seeing the bottom 4 stops of shadow detail squeezed into just 4 bits of data by the GH4's 8-bit H.264 encoder. Such a crude level of digitization seriously undermines the practical value of extending the dynamic range to 12 stops. With a 10-bit encoder, the 4-stop shadow range is encoded at 6-bit precision, literally four times as much image detail. That is why using an external recorder with 10 bits or more precision should make a significant improvement in image quality over the internal encoder.
Barry_Green
Moderator
I think this issue is being overthought and overblown. Consider that a Canon C300 uses an 8-bit log curve, and if you "pull" the C300 to 320 ISO, it delivers the same 12 stops as VLOG-L, in the same proportions (8 stops below middle gray, 4 stops above), and it does so in 8-bit. And it can't even output 10-bit; its SDI is limited to 8-bit...
CLOG is slightly different from VLOG in a couple of ways; CLOG is optimized to deliver 7 stops below and 5 stops above middle gray, VLOG renders 8 stops below and 4 stops above. However, VLOG renders middle gray at 42 IRE, CLOG renders it at 33 IRE. So CLOG has to fit 7 stops in the range of 32-87 (in 8-bit values).
CLOG 8-bit renders 0% reflectance at code value 32; VLOG renders it at the exact same spot, 32. In both, that represents 7.3 IRE. They are the same.
CLOG renders 18% reflectance at code value 87, at 32.8 IRE; VLOG renders it at code value 108, for 41.7 IRE. So VLOG actually renders the region from black to middle gray with more precision (76 shades) than CLOG does (55 shades). But VLOG is fitting one more stop in that range, 8 stops below middle gray, vs. CLOG's 7 stops. Unless you're pulling CLOG down to 320 ISO, in which case it will be fitting 8 stops under middle gray into that range. Now, if you look at the VLOG-L curve, it renders the first stop below middle gray (where Canon renders its middle gray) at about code value 89 ... pretty much exactly the same as CLOG.
CLOG renders 90% reflectance at code value 153 (62.7 IRE); VLOG renders it at 151 (61 IRE). Code 151 hits at about stop #10 on VLOG-L, and about the same place on CLOG. In fact, given the performance in the lower stops, and that 90% hits at almost identical spots, I would say that CLOG and VLOG are pretty much identical from pitch black up to 90% reflectance, the first 10 stops of exposure.
On the Canon chart, they show the first 4 stops of CLOG are encoded in code values from 32 to 48 -- 16 steps. Exactly the same as VLOG-L does.
So is the VLOG-L somehow crippled and can we expect awful results? We should only have to look to the C300's CLOG for that answer; do people complain about CLOG and say that its lower stops are unusable? (I'm asking, I don't know, I haven't delved into the C300 line at all).
In the lower range, CLOG and VLOG-L are both 8-bit log gammas that render their bottom 10 stops pretty much exactly the same.
CLOG renders maximum peak clipping at code 254, whereas VLOG-L does it at about 180. That's the only place that there's a compromise, near as I can tell, is that it doesn't use the full range of code values from 181 to 255 the way that CLOG does. Even so, the stops that it does use are rendered proportionately exactly the same, as LOG is intended to do; the curve is pretty much exactly equidistant for the top 8 stops of the range (4 stops below to 4 stops above middle gray).
So if there's a compromise in VLOG, it's that the very top two stops don't use as many code values as they otherwise might could have. In pretty much every other way, it looks like it's mapped largely the same as Canon's CLOG. Unlike CLOG however, in the GH4 and the DVX200 you do have the option to output 10 bits and thus take advantage of 4x as many shades.
(none of this is to say that 10-bit wouldn't be better; of course 10-bit would be better... that's obvious. But I don't think it's a reasonable conclusion to say that VLOG-L is somehow artificially hampered, when it's delivering largely the same allocation and same performance as Canon's CLOG on their C300).
CLOG is slightly different from VLOG in a couple of ways; CLOG is optimized to deliver 7 stops below and 5 stops above middle gray, VLOG renders 8 stops below and 4 stops above. However, VLOG renders middle gray at 42 IRE, CLOG renders it at 33 IRE. So CLOG has to fit 7 stops in the range of 32-87 (in 8-bit values).
CLOG 8-bit renders 0% reflectance at code value 32; VLOG renders it at the exact same spot, 32. In both, that represents 7.3 IRE. They are the same.
CLOG renders 18% reflectance at code value 87, at 32.8 IRE; VLOG renders it at code value 108, for 41.7 IRE. So VLOG actually renders the region from black to middle gray with more precision (76 shades) than CLOG does (55 shades). But VLOG is fitting one more stop in that range, 8 stops below middle gray, vs. CLOG's 7 stops. Unless you're pulling CLOG down to 320 ISO, in which case it will be fitting 8 stops under middle gray into that range. Now, if you look at the VLOG-L curve, it renders the first stop below middle gray (where Canon renders its middle gray) at about code value 89 ... pretty much exactly the same as CLOG.
CLOG renders 90% reflectance at code value 153 (62.7 IRE); VLOG renders it at 151 (61 IRE). Code 151 hits at about stop #10 on VLOG-L, and about the same place on CLOG. In fact, given the performance in the lower stops, and that 90% hits at almost identical spots, I would say that CLOG and VLOG are pretty much identical from pitch black up to 90% reflectance, the first 10 stops of exposure.
On the Canon chart, they show the first 4 stops of CLOG are encoded in code values from 32 to 48 -- 16 steps. Exactly the same as VLOG-L does.
So is the VLOG-L somehow crippled and can we expect awful results? We should only have to look to the C300's CLOG for that answer; do people complain about CLOG and say that its lower stops are unusable? (I'm asking, I don't know, I haven't delved into the C300 line at all).
In the lower range, CLOG and VLOG-L are both 8-bit log gammas that render their bottom 10 stops pretty much exactly the same.
CLOG renders maximum peak clipping at code 254, whereas VLOG-L does it at about 180. That's the only place that there's a compromise, near as I can tell, is that it doesn't use the full range of code values from 181 to 255 the way that CLOG does. Even so, the stops that it does use are rendered proportionately exactly the same, as LOG is intended to do; the curve is pretty much exactly equidistant for the top 8 stops of the range (4 stops below to 4 stops above middle gray).
So if there's a compromise in VLOG, it's that the very top two stops don't use as many code values as they otherwise might could have. In pretty much every other way, it looks like it's mapped largely the same as Canon's CLOG. Unlike CLOG however, in the GH4 and the DVX200 you do have the option to output 10 bits and thus take advantage of 4x as many shades.
(none of this is to say that 10-bit wouldn't be better; of course 10-bit would be better... that's obvious. But I don't think it's a reasonable conclusion to say that VLOG-L is somehow artificially hampered, when it's delivering largely the same allocation and same performance as Canon's CLOG on their C300).
Last edited:
Thanks for detailing the similarities and differences between Panasonic's V-Log-L and Canon's CLOG. While the argument that "if it's good enough for Canon..." shows that an 8-bit log profile video camera can be successfully marketed, I think it says far more about Canon's marketing clout than the camera's image quality. As your analysis of V-Log-L shows, Panasonic's implementation on the DVX200 (and presumably the GH4) squeezes 12-stops of dynamic range into significantly less than 8-bits of data, producing results closer to 7-bit precision. I examined a wide variety of Panasonic shadow detail samples in the course of my high-bitrate work on the GH2, and I saw many cases where inadequate bit-depth encoding produced visibly degraded images. There are a lot areas in videography that are genuinely subjective, but there's only so far you can push things when you just don't have enough bits...
Barry_Green
Moderator
What I'm saying is -- speculation based on the numbers may or may not prove fruitful; the results derived from actually using the product will be far more illustrative. I don't know how CLOG has been received, but I point it out because it is fundamentally the same thing, and it's been on the market for, what, five years now? That leads me to the conclusion that if it hasn't been declared a total disaster by the users, then maybe VLOG shouldn't be either -- at least until people get it in their hands and actually use it.
Saying that it's "significantly less than 8 bits of data" is -- misleading, don't you think? You can't represent the full tonal scale of VLOG-L in any less than 8 bits. 7 bits won't fit it all. Granted it doesn't use the full range provided by 8 bits, but that doesn't mean that it's fair to say it's "significantly less than 8-bits of data".
Even if it was -- that same argument can be taken to logical extremes to declare that all 8-bit cameras are useless doorstops when compared to the magnificence that is 10-bit; 10-bit provides so much more data than 8-bit that 8-bit becomes laughable, right? And yet -- 8-bit cameras have been the stalwart and staple of professional video production since digital video was invented; 10-bit recording has only really become commonplace quite recently. Every frame of DVCPRO-HD or HDCAM that's ever been shot, was done so in 8-bit. The Star Wars prequels (II and III) were shot on 8-bit cameras. Every episode of "Planet Earth" was shot on 8-bit. Every frame of HDTV that anyone watches, broadcast, is done so in 8-bit.
So yes, 8-bit would be an easy target to disparage, but -- in reality, it does work really quite well. 10-bit is better, sure, but it is not the orders of magnitude better that the sheer numbers might make you think -- especially when you consider that the presence of noise can limit or even eliminate the advantages offered by 10-bit encoding.
Which leads me back to -- the images will do the talking. Those who have used VLOG on a GH4 sure do seem to like it generally. Perhaps the results prove more optimistic than a simple examination of the numbers might indicate?
Saying that it's "significantly less than 8 bits of data" is -- misleading, don't you think? You can't represent the full tonal scale of VLOG-L in any less than 8 bits. 7 bits won't fit it all. Granted it doesn't use the full range provided by 8 bits, but that doesn't mean that it's fair to say it's "significantly less than 8-bits of data".
Even if it was -- that same argument can be taken to logical extremes to declare that all 8-bit cameras are useless doorstops when compared to the magnificence that is 10-bit; 10-bit provides so much more data than 8-bit that 8-bit becomes laughable, right? And yet -- 8-bit cameras have been the stalwart and staple of professional video production since digital video was invented; 10-bit recording has only really become commonplace quite recently. Every frame of DVCPRO-HD or HDCAM that's ever been shot, was done so in 8-bit. The Star Wars prequels (II and III) were shot on 8-bit cameras. Every episode of "Planet Earth" was shot on 8-bit. Every frame of HDTV that anyone watches, broadcast, is done so in 8-bit.
So yes, 8-bit would be an easy target to disparage, but -- in reality, it does work really quite well. 10-bit is better, sure, but it is not the orders of magnitude better that the sheer numbers might make you think -- especially when you consider that the presence of noise can limit or even eliminate the advantages offered by 10-bit encoding.
Which leads me back to -- the images will do the talking. Those who have used VLOG on a GH4 sure do seem to like it generally. Perhaps the results prove more optimistic than a simple examination of the numbers might indicate?
Last edited:
joe12south
Veteran
Thank you, Barry. The computer programmer in me is offended that they don't use every precious bit they can, but the filmmaker in me accepts that if you'r going to compromise, do it in the highlights where we can't see well anyhow.
dustylense
Veteran
I don't take pictures and moving images based of press release, white paper, or speculation. I take them by turning the camera on and shooting. So while everyone is trying to out smart each other without anyone having used the darn firmware (except maybe Barry), how about we hold our horses for a week or two and see how this darn vlog works in the hands of SHOOTERS!
Sorry, but the technical details you documented on Panasonic's V-Log-L implementation are not "speculation", they are hard specifications that impose rigid limits on the bit-depth of image details that can be preserved by an 8-bit H.264 encoder. Will people still use it with the GH4 encoder anyway and manage to produce pleasing subjective results? Of course, but it will demonstrate nothing more than the power of confirmation bias to vindicate any amount of marketing hype.
I don't think either of us are resorting to "logical extremes", we're talking about the realistic limits of high quality video data crunching. And no, it is not misleading to point out that V-Log-L is designed and specified to produce "significantly less than 8 bits of data" when used with an 8-bit H.264 encoder. As you have detailed, it uses only about 150 code levels to encode 12-stops of dynamic range. That is much closer to 7-bit precision (128 levels) than it is to 8-bit precision (256 levels). Based on my experience as a DSP engineer, that is not quite enough to avoid visible degradation, but if you expose carefully, no one will notice.
So yes, in most circumstances, 8-bit H.264 encoding is good enough to produce high quality results, as long as you have enough data bits to handle the dynamic and tonal ranges of the subject matter without visible degradation. But packing 12-stops of log data into H.264 macroblocks is not a typical situation, it demands far more technical precision than the 10-stop Rec 709 profiles the GH4 was originally built to use. I think Panasonic's decision to prioritize Varicam LUT compatibility at the expense of 8-bit image quality clearly shows that they have designed this upgrade explicitly to serve the needs of cinematographers using 10-bit or better external recorders.
Last edited:
dustylense
Veteran
who cares???? Really. Who here gives a flying F? You're talking about a $1400 prosumer camera. We shoot out cats, some trees, ourselves making coffee and once in a while, maybe a corporate video for YouTube.
Tonight, I'm gonna tell my wife that the shadows on the gh4 only have 4 bits and totally ruin my chances for nookie time.