A7s and 8bit banding

[llirik]

Interesting topic. And clearly some very knowledgable people here, so thought I'd ask.

There was an issue I was having some time back, and it hasn't really reappeared. At least not noticeably.
But when I first picked up the shogun and was recording from a7s, using my standard luts and shooting in slog2, i had this kind of banding happening:



At the time, I simply figured slog2 isn't good for indoors and to use it outdoors only. But there is some curious behaviour going on... if you pay attention.. the skin tones are fairly smooth and fine... but some colors - like the grey on his suit, or in his hair are quite severe.

Any ideas on why this is only happening in some places and not others? Footage may have been slightly overexposed by 1 stop.

And it's almost always on parts of the frame that are out of focus. Stuff that is in focus is fairly good, only minor (you can see on the white uniform)

 

[paulcurtis]

Interesting topic. And clearly some very knowledgable people here, so thought I'd ask.

There was an issue I was having some time back, and it hasn't really reappeared. At least not noticeably.
But when I first picked up the shogun and was recording from a7s, using my standard luts and shooting in slog2, i had this kind of banding happening:

View attachment 102692View attachment 102693

At the time, I simply figured slog2 isn't good for indoors and to use it outdoors only. But there is some curious behaviour going on... if you pay attention.. the skin tones are fairly smooth and fine... but some colors - like the grey on his suit, or in his hair are quite severe.

Any ideas on why this is only happening in some places and not others? Footage may have been slightly overexposed by 1 stop.

And it's almost always on parts of the frame that are out of focus. Stuff that is in focus is fairly good, only minor (you can see on the white uniform)

Is that source footage untouched. Because this is exactly what happens when i've seen some LUTs added, banding in the out of focus areas especially (as they're usually subtle gradients)

What was the post route for these images?

cheers
Paul

 

[paulcurtis]

i don't know if i'd count less dynamic range as a plus (unless i misunderstood what DR was intended to mean), and our version of slog2 was likely designed with 8bit in mind (as well) as it had to be designed specifically for the a7s (evidenced i suppose by how it reacts differently than slog2 on other sony cameras).
as for smarter distribution, i think that's a personal preference (and i'll cede that your preference/opinion might actually be more correct or in-line with the industry than mine). you like it saved in the shadows, but im fine with the dr of slog2 protecting the highlights and then later boosting shadow and denoising a bit. even with a super (several stops) underexposed face (like in the link that you shared from my recent shoot), clearly something is lost, but it'll boost back up a stop or two pretty easily i think and still look fine (as an underexposed face... obviously if i was trying to bring it up to 60ire or something it'd be a lost cause).
the two 'ands' though that i agree with are indeed big ones, and a good reason the c300 is in a different league and price point than the a7s

I've mentioned this a few times and it's worth repeating

Yes, this is 'our' Slog2 curve, unique to the A7s. The darks are darker for example and we need to pay very special attention to highlights, especially if we're recording ProRes externally.

(The reason is prores is supposed to always be legal levels, but actually the PR that the Odyssey produces is full range and lots of apps do not expect that, hence immediate clipping of highlights)

If you pass these images through a LUT which is expecting a starting point of normal Slog2, you will clip top and bottom, if you're going to use a LUT you have to make sure it knows the source curve or you need to manipulate your curve into the same range the LUT is expecting.

cheers
Paul

 

[llirik]

Is that source footage untouched. Because this is exactly what happens when i've seen some LUTs added, banding in the out of focus areas especially (as they're usually subtle gradients)

What was the post route for these images?

cheers
Paul

Yeah - this is with a LUT applied. I don't recall the raw prores, but I think it did not have this issue (project's been archived, would need to pull it up).

Thing is - i use these luts other times on the same footage process, but without the issue. It was particularly bad on this one event, and a few random times pops up. So i'm not sure of exactly why it's happening.
It's never happened on s700 slog2.

Recorded a7s slog2 to shogun in 4k prores HQ. Dumped and edited in FCPX using LUTility.

 

[paulcurtis]

Yeah - this is with a LUT applied. I don't recall the raw prores, but I think it did not have this issue (project's been archived, would need to pull it up).

Thing is - i use these luts other times on the same footage process, but without the issue. It was particularly bad on this one event, and a few random times pops up. So i'm not sure of exactly why it's happening.
It's never happened on s700 slog2.

Recorded a7s slog2 to shogun in 4k prores HQ. Dumped and edited in FCPX using LUTility.

There are many types of LUTs, they work by mapping an R,G,B combination to another R,G,B result. They're finite though, it's not a analogue mathematical mapping but a look up. The size of a LUT dictates how many possible combinations there are, 8x8x8 to 64x64x64 and up. However when dealing with finely graduated areas you may find that the LUT in that particular area of colour is not very accurate or dense and two similar source colours become the same destination colour - banded.

The application applying the LUT also has work to do, because it ought to be interpolating between RGB points that aren't there by looking at the neighbouring points. But there are many ways to interpret, from linear, cubic and so on.

Therefore it is entirely possible that the LUTs you're using a not good enough in terms of accuracy or the application is doing a poor job of interpolating. Or a combination.

LUTs are useful, they're great for mapping non linear complex transforms (like film print emulation) but they have to be made right.

cheers
Paul

 

[caseywilsondp]

Which luts were using, or does lutility have them for slog2 (I've never used it so I'm not sure exactly what it is)?

 

[Shirozina]

I'm finding with the controls available in Resolve I can get an acceptable looking image without a LUT and without banding in Slog2 / S-Gamut. It's a fine balancing act but it's possible and easier to do than in Premier. Still not convinced the overall image quality is anywhere near as good as that from the BMPCC though. I can get the colors to look OK but the overall image just looks plastic and artificial. I just don't think it can reveal subtle tonal and hue variations that are needed to create a sophisticated 'cinematic' image...........................

 

[Razz16mm]

Interesting topic. And clearly some very knowledgable people here, so thought I'd ask.

There was an issue I was having some time back, and it hasn't really reappeared. At least not noticeably.
But when I first picked up the shogun and was recording from a7s, using my standard luts and shooting in slog2, i had this kind of banding happening:

View attachment 102692View attachment 102693

At the time, I simply figured slog2 isn't good for indoors and to use it outdoors only. But there is some curious behaviour going on... if you pay attention.. the skin tones are fairly smooth and fine... but some colors - like the grey on his suit, or in his hair are quite severe.

Any ideas on why this is only happening in some places and not others? Footage may have been slightly overexposed by 1 stop.

And it's almost always on parts of the frame that are out of focus. Stuff that is in focus is fairly good, only minor (you can see on the white uniform)

Some CMOS sensors will exhibit horizontal streaking from hot specular light sources similar to vertical streaking on CCD's.

 

[Razz16mm]

i think i'd agree with you, but then again slog2 is shooting at a minimum iso of 3200 with no internal denoising, and again, at 8 bit, so there definitely is a compromise. i think what'd be interesting is testing a scene with your bmpcc and downscaling 4k slog2 footage and seeing if it felt comparable in 'cleanness' or not.

The major compromise is high levels of compression. 8 bit uncompressed 4:2:2 broadcast YUV at 800 Mbps doesn't show banding. 4:2:0 is very prone to banding at any bit rate common to low end cameras. Log curves are value compression that flattens contrast and gamma values into a very narrow DR range to preserve detail that would otherwise fall outside the linear DR range of a video codec.

 

[Shirozina]

The major compromise is high levels of compression. 8 bit uncompressed 4:2:2 broadcast YUV at 800 Mbps doesn't show banding. 4:2:0 is very prone to banding at any bit rate common to low end cameras. Log curves are value compression that flattens contrast and gamma values into a very narrow DR range to preserve detail that would otherwise fall outside the linear DR range of a video codec.

Wouldn't the uncompressed 8bit 4.2.2 available from the HDMI port offer some advantages in this respect over the internal codec?

 

[Emanuel]

sample extracted from (minute 28:30):
https://vimeo.com/114978513

 

[MichaelFischer.at]

Nothing to show but I just tried to 'break' some test footage in PP and Resolve - SLOG / S-Gamut and Picture profile off. It's IS harder to break it in Resolve. With picture profile off ( Portrait setting with contrast and sharpness off) I was getting more banding than SLOG in a graduated evening blue sky.

Besides shooting S-Log2, I also like to use PP off, STD -3,0,-2. This is close to the gamma of BMDFilm4k and I'm able to linearize that quite well. While shooting a test scene with myself as a subject I noticed a similar thing. There was banding (or a much worse color separation) with the "STD-3" footage while the S-Log2/Movie footage looked good.
FYI, this was done in Resolve with the same curves/process after linearization. Don't have examples right now, but can provide them later on if you're interested!

 

[cpc]

The Creative Styles aren't really supposed to have grading-grade (pun intended) color separation. They are all meant to be final looks.
Could be the colors are too compressed or the Contrast parameter is applied on the 8-bit values, hence bandng when tweaking in post.

 

[Shirozina]

This is all getting rather interesting......

 

[paulcurtis]

sample extracted from (minute 28:30):
https://vimeo.com/114978513

Wow, i'd suggest something was up with your workflow for the 4:2:0 version, that looks so far away from what it should be. I know nothing about the GH4, but still seems unusually bad...

cheers
Paul

 

[paulcurtis]

The major compromise is high levels of compression. 8 bit uncompressed 4:2:2 broadcast YUV at 800 Mbps doesn't show banding. 4:2:0 is very prone to banding at any bit rate common to low end cameras. Log curves are value compression that flattens contrast and gamma values into a very narrow DR range to preserve detail that would otherwise fall outside the linear DR range of a video codec.

Generally this is all correct, but i would add

this is 8 bit in YUV, not RGB. So those Luma and difference channels are combined to produce the RGB, which means so long as that combining happens in a suitably large colour depth then banding is even more minimised. It's not like popping photoshop into 8 bit mode. You cannot say 256 levels of R,G and B.

Same deal with RAW on the FS700, being 11/12 bit RAW pre white balance means after white balancing than it's a far greater range.

I wrote about some aspects of this here

http://blog.inventome.com/Blog/2015...d-Noise-Workflow-with-UHD-and-Odyssey-7Q-Plus

I don't have much time to write but i have a raft of other tests that i've done i'd like to write up but the one above tries to break the banding.

cheers
Paul

 

[Shirozina]

A lot of interesting info on your blog Paul - thanks. Looks like we are dealing with poor internal image resampling as well so I think the only obvious thing to do now is to stop messing about and get the Shogun external recorder.

 

[benjaminreece]

Before you all go buying shoguns- see this thread- Premiere (have not tested the new 2015 update) has problems with XAVC- specifically the video engine does not access the complete color space of the codec. Actually Nate Weaver explains it best: http://www.dvxuser.com/V6/showthread.php?337008-Severe-Banding-in-S-amp-Q-mode-180FPS/page2

 

[Samuel H]

Wow, i'd suggest something was up with your workflow for the 4:2:0 version, that looks so far away from what it should be.

Those samples are procesed: there's a ton of contrast added to them. You'd be hard pressed to see the banding in the original footage. You can try, actually, he posted the link where you can find it!!

 

[Emanuel]

Those samples are procesed: there's a ton of contrast added to them. You'd be hard pressed to see the banding in the original footage. You can try, actually, he posted the link where you can find it!!

Yes, and I am only the messenger, not my testing actually, so be gentle we the messengers are always innocent people... : D

You can somehow download the video, it is worthy of watching

sample extracted from (minute 28:30):
https://vimeo.com/114978513