I can always spot a VLog L clip, no matter how it's graded.
GH5 GH5s vs GH4 dynamic range in Cine D
- Thread starter RedChalice
- Start date
Sage
Well-known member
Yes, it is a VLog clip. If I understand correctly, the implication is that GHa is a color grade, rather than a color science conversion (indicated by this image)?
I mean to understand better, forgive me if this is mistaken.
What's obvious from ungraded VLog L footage is that it starts out with extremely low contrast and saturation. That gives it a bleached, milky cast that's difficult to completely eliminate. This makes VLog L best suited for muted color palettes and Scandinavian scenery. The technical issue is that even with 10-bit color, VLog L crushes microcontrast and saturation bit-depth in order to maximize overall dynamic range. That gives it detailed highlights, but the bit-depth of those details is impoverished, particularly in color discrimination. So when you attempt to enhance contrast and saturation in your color grade, you can only go so far before it starts to look blotchy and the colors bloom. Check out the hue patches in the color chart the girl on the right is holding in the screen grab above. The pure colors have an overamped neon glow even though the girls' complexions still look milky. OTOH, the intense reflection of the LED light in the background looks sliky smooth.
Cary Knoop
Veteran
Sheer nonsense!
stoneinapond
Veteran
Whether that is nonsense or not, it definitely helps explains the issue I've had when trying to grade VLog-L to my satisfaction.
So I'll go with that.
Thanks Lpowell
Cary Knoop
Veteran
10 bit V-Log-L is sufficient to handle the GH5 dynamic range, it grades just fine.
I have to agree and furthermore the colors on those conversions are (and I'm being kind here) - not what I would want to end up with....
I've been doing some more testing of HLG vs V-LOG and with a bit of work in Resolve I can get them to match pretty well and certainly to the extent you couldn't tell which started out as what.
stoneinapond
Veteran
Testing and getting something that satisfies a creative vision are two different things. Emphasis here on "individual satisfaction."
There are subtle dimensions of grading quality that are hidden beneath the concept of "dynamic range". I think we'd all agree that the constricted DR of Rec 709 crushes both highlight and shadow details, with an S-shaped tone curve that confines them to the extreme ends of the encoded range of data. Vlog L flattens that tone curve, extending its logarithmic range to capture significantly more details in both highlight and shadow regions. That gives you more lattitude in grading the highlights and shadows of high-contrast images.
But consider how the trade-offs in each case affect the quality of the recorded visual data. With Rec 709, most of the H.264 encoding range is reserved for the midtones. That allows it to encode fine distinctions in skin tone and color saturation, but only within the central region of the tone curve. With Vlog L, the H.264 encoding range is spread over a much wider dynamic range. Consequently, it must devote more bits to encoding highlights and shadows, and fewer to midtones. Even with the bit-depth boost to 10-bit encoding, VLog L uses less than 80% of the IRE range, reducing it to an effective encoding range of about 9-bits.
You can see with your own eyes how VLog L manages its available H.264 encoding bandwidth. An elevated pedestal bumps the darkest shadow regions into low-contrast shades of gray, and highly saturated colors are compressed into pastel shades. That gives the footage a broad dynamic range to encode the extremes of shadows and highlights, but with relatively limited ability to distinguish subtle variations in midtone details. To do so with perceptibly flawless precision, it would need even more bit-depth to capture the fine gradients in near-gray and flesh tone hues. Unfortunately, that is where H.264 drops the ball, with the degraded resolution and bit-depth of its half-res chroma channels.
A good grade with a proper conversion LUT can ameliorate most of these issues. We're used to seeing color shades of objects vary in different lighting conditions, which allows us to grade across a broad palette of natural-looking shades. But our eyes are highly sensitive to the coloration of certain well-known objects, such as the sky and human complexion. When these details are encoded with coarse precision, the result is often seen as banded near-gray sky gradients and milky or waxy skin tones. While you can always give your subjects more light, you can't always give them more bits.
Cary Knoop
Veteran
It's all very nicely written, but no, I disagree.
Your basic argument is never use log, any log even if it is 10 bit, it's not good enough, just use Rec.709.
It seems similar to your arguments against 4:2:2 you made in another topic, that 4:2:0 is just fine or your language might have even suggested it's better than 4:2:2.
It looks to me that you invested psychologically in Rec.709 4:2:0 technology and think the rest is not neccesaary.
For the GH5 I actually do not use V-Log-L very often, I mostly use HLG but certainly not for reasons that a 10 bit encoding is insufficient for V-Log-L.
By the way your suggestion that the extreme ranges get more discrete levels is not correct, a log OETF does indeed increase the available discrete shadow levels however it actually decreases the available discrete highlight levels.
Last edited:
LOL, that's quite a strawman you've conjured up. If your grading skills are even half as impressive, I've no doubt you can craft a masterpiece, VLog L not withstanding.
But nope, I most often shoot with 4:2:2 color in log gamma, just not with VLog L. There are a number of cameras with log profiles, and not all of them produce as bleached a look as VLog L. They do, however, all share a similar weakness in microcontrast, as much due to H.264 compression as the log profile. What I do to compensate is to shoot in 4K resolution and resample down to 1080p. That produces HD footage that is effectively 4:4:4 color with two additional bits of precison.
Cary Knoop
Veteran
Somehow I expected that preserving 4k resolution was not one of your priorities
Reducing 4K to 2K increases the luminance bit-depth by two, while for 4:2:2, unlike 4:2:0 the color bit-depth increases a smaller amount as well.
Last edited:
Cary Knoop
Veteran
Indeed, 4:2:0 becomes 4:4:4. however the bit-depth for 4:2:0 remains the same while for 4:2:2 it increases. The luma channel's bit-depth increases by 2.
So a 10-bit 4K 4:2:0 source properly downscaled to 2K becomes a 12-bit luma and 10-bit chroma 4:4:4 source.
Factually the chroma resolution simply stays the same during the downscaling, in 4K 4:2:0 Y'CbCr the chroma resolution is already 2k, all you need to do is to downscale the luma channel, with the advantage that the luma values get 4x over sampled which effectively increases the bitdepth by 2.
Last edited:
Sage
Well-known member
Thanks for explaining that, I understand where you're coming from
By the way, huge fan here. I think your Flow Motion v2 patch was a technical masterpiece (unparalleled balance of reliability and quality from Day 1). Its been on my GH2 for years
Its late (just got back from a Trombone Shorty concert); tomorrow, 4k!
Sage
Well-known member
Its been getting lost in the mix; that image was an example of *problematic color
This is incorrect color:
This is correct color:
You can get an HLG to VLog conversion here. It was done with the very same 25k sample interpolation as GHa; 2.2 multipass & smoothing update soon
Its been getting lost in the mix; that image was an example of *problematic color
This is incorrect color:
This is correct color:
You can get an HLG to VLog conversion here. It was done with the very same 25k sample interpolation as GHa; 2.2 multipass & smoothing update soon
The skintones look terrible in your 'correct color' image. They may 'measure' correctly but it looks wrong. Also the overall image appearance looks like what happens when part of the tone curve is reversed or flattened out. I'm intrigued by the concept and can only applaud the effort you have put in but on the basis of what I have seen so far I'm not parting with my money.
Some other misconceptions floating around here. V-LOG looks 'milky' when viewed without a LUT so it will retain this look when processed as data is crushed or lost. No it won't as all you are seeing when you view a log image on a REC709 device is the video card and screen's visual interpretation of the LOG image and not the data contents of the image file. A LOG file (esp one captured in 10 bit 4.2.2 with a good bitrate) contains all the colour and tonal information and more that is needed to create whatever you want. The only occasion you would struggle to reconstruct a good image is if you took a screen grab of the log image (which BTW is what the corrected samples look like) and tried to work with that or captured or transcoded the log file with an 8bit 4.2.0 low bitrate codec. Also downsampling the image size does theoretically increase the chroma subsampling but it's not recovering lost color information and is just filling in the gaps with interpolated data which is not the same as capturing it in the first place.
Sage
Well-known member
This is intriguing! That is the Alexa Mini, with a Master Prime 35mm
I think I understand where you're coming from though. You're part of a select group (the 5th person I know of) that has a strong aversion to logarithmic luma. I know what it feels like, as I have the same aversion to linear
In these cases, I recommend the LogC+Rec709 route (one of the conversion variations is just to raw LogC). I am also thinking to release a linear variant of EC
Alexa Mini LogC
Alexa Mini LogC + Rec709 (Linear)
This is intriguing! That is the Alexa Mini, with a Master Prime 35mm
I think I understand where you're coming from though. You're part of a select group (the 5th person I know of) that has a strong aversion to logarithmic luma. I know what it feels like, as I have the same aversion to linear
In these cases, I recommend the LogC+Rec709 route (one of the conversion variations is just to raw LogC). I am also thinking to release a linear variant of EC
Alexa Mini LogC
Alexa Mini LogC + Rec709 (Linear)
Yes much, much better (to my eyes anyway)!