FS7: Does Resolve Read the Metadata EI?

D

digitalinnovations

Well-known member
I have some footage that was shot in CineEI and the exposure index switch was flipped to 800 instead of 2000. So my question is: does Resolve read the exposure index metadata? Is there a way to either change or ignore this in Resolve? I just don't want to have to push the footage anymore than I need to in post.
 
let's hope in the next sub version release. We should head over to blackmagic forums and make sure we ask.. those guys are great, and very responsive in my experience.
 
Jeremey,

if you've shot ISO 800 when you thought you were shooting ISO 2000, you've actually pulled the exposure (overexposed). Provided you've not clipped anything you'll need, absolutely no harm done and your shadows should end up a bit quieter.

If you just want to apply an S-Log2 or S-Log3 exposure shift, it's pretty easy to create a 1D LUT for that. I've written an app for that! ;-)

Just to save any effort now, I've just knocked up a couple of 1D cubes (for S-Log2 and S-Log3). Dropbox link:
F5 CineEI ISO 800 Exposure Corrections

Just bung the .cubes into the Resolve LUT directory, rescan it (or start Resolve) and provided the log clips have been read as data range (auto should do that, but no harm in being explicit) you can use the LUTs to shift your hot S-Log2 or S-Log3 to correct exposure,

Ben
 
digitalinnovations said:
I just don't want to have to push the footage anymore than I need to in post.
Click to expand...

Even if it did read it, it would pretty much mean exactly the same in terms of how the exposure would have to be compensated in post, the only difference is the compensation would happen automatically. And as Cameramanben said, shooting with EI rating of 800 means overexposing, so no pushing but pulling. And of course if you just flipped the EI rating switch from 2000 to 800 but did not react to the change in exposure as simulated with LUT by opening the iris, adding light, etc. (because perhaps you were not using any LUT ) then how the Slog was recorded did not change at all.
 
Last edited:
Ben, is there any visual benefit to applying a 1D LUT for exposure vs compensating for exposure manually with node #1 and then grading with subsequent nodes after that? In other words, is a LUT actually better than manual or just a different approach?
 
Hmm Random thought as well workflow-wise.....

In my situation I shoot for several companies, some like LOG, some like baked in... I find some of them know some stuff... other know very little. For those who like LOG, I provide a LUT ...etc... This doesn't respect the EI I've chosen in camera though (unless I'm providing LUTs for every EI I've used and slated each shot that way).. so unless I give them the info or let them know, I'm left trusting they will colour it correctly etc.

What if...
Using LUTCal... I create an Slog3 SGamut3.cine EI 2000 record gamma.....
Change nothing....
And leave the output gamma as Slog3 SGamut3.cine as well

Wouldn't this technically allow me to turn internal record on, and be able to hand over Slog3 footage that (in a backwards way) respects how I've rated the camera? Allowing someone to drop a LUT on without having to worry about exposure?
Sure it might mean I'm stuck looking @ LOG on the VF... but I can fix that with an external monitor w/LUT applied there....

Thinking out loud mostly... but something I'm going to have a play with to see how it behaves!
 
Hi Doug,

I'm not a huge user of Resolve, but as far as I'm aware the exposure tools available are lift / gamma / gain for linear or conventional gamma material and shadow / midtone / highlight offsets as log adjustments. Technically an exposure adjustment on log material is a matter of offsets, in exactly the way that it is gain on linear material; hence Resolve's offset options under 'log'.

What is fiddly is that for any log recording the amount of offset varies - above mid gray it tends towards a constant, but heading downwards it shrinks. The adjustment is actually pretty simple to do as a bit of code if you know the parameters - the LUTs I posted just do what the ISO slider does in Sony software which reads the metadata. Effectively it's log to pure linear, linear gain, back to log.

There are about half a dozen well documented flavours of log in common use and three or four less well documented (ahem, BMD Film!). Putting aside reading the metadata automatically, I honestly don't get why simple log ISO sliders don't seem to crop up in post software. Ah well,

Ben
 
Last edited:
Thanks everyone.

Ben: that was super nice of you to take the time to do that. I'm on my phone now but will grab those. I do have your app purchased through the App Store and love reading your posts. Always informative even if I don't quite understand it all yet :)

Currently I do not use a monitor LUT (may be why I didn't notice it till after the fact) so I did not adjust exposure based on the rating and avoided clipping by monitoring the waveform.

I downloaded Catalyst Browse out of curiosity and since it reads the metadata I could see what the difference between ISO 800 and 2000 would've been.
 
chappelldigital said:
Hmm Random thought as well workflow-wise.....
I create an Slog3 SGamut3.cine EI 2000 record gamma.....
Change nothing....
And leave the output gamma as Slog3 SGamut3.cine as well

Wouldn't this technically allow me to turn internal record on, and be able to hand over Slog3 footage that (in a backwards way) respects how I've rated the camera? Allowing someone to drop a LUT on without having to worry about exposure?
Click to expand...

If you are saying, can you just bake-in the CineEI ISO to S-Log2/3 material, there's no need to make a LUT for that, just select S-Log3 from the built-in options. BUT.....

Sony is very strict about the clip level of log recording. In the case of S-Log2, that's the top of extended range anyway, but in S-Log3 that is around 94% IRE, and if you try shifting the ISO up you'lll find it's still around 94% IRE. In other words if you do push to a higher ISO, you will lose dynamic range from the highlights. If you pull to a lower ISO, you reduce your headroom; with normal CineEI technically you still have the same dynamic range, just more shadow detail. If you bake in, you will be cutting that off.

Baking in it becomes just like the C300 - full dynamic range at native ISO, but each stop you go either way is a stop lost. You are starting off with quite a few stops, so I'm not saying don't, but I think I'd do a chunk of testing and have a good chat with a colourist before handing it over on paid work!

Ben
 
Hi Ben,

I appreciate the honesty in your answer about the exposure LUT vs. Manual. Since it sounded like maybe you weren't quite, sure I decided to do some head to head testing. Now, I don't shoot S-LOG3 very often and I don't purposely over-expose to 800 EI, but coincidentally I did shoot S-LOG3 at an airshow last week as an experiment and (not surprisingly) there are some over-exposed shots which made good testing matieral. But of course, granted it's not faces which is what matters most to most people.

So I downloaded the two 1D LUTs you provided in an earlier post, took a few of those airplane shots into Resolve, and tinkered around with the exposure a few different ways -- with the final node being the TypeA LUT.

Here are my conclusions:

1) Your LUTs work as good as manually adjusting for over-exposure -- but no better than manual. The two methods can be made to look identical in about 10 seconds with a split screen wipe.

2) Your LUT is still going to require some manual adjustment if the exposure wasn't perfectly at 800 EI. So, if some manual adjustment is probably going to be needed anyway, I'd just skip the LUT and go right to manual adjustment. But your LUT would be great for dailies or something quick like that.

3) Making changes with the Primary Wheels or Primary Bars provides much better results and is a whole lot easier than using the LOG wheels. I already knew this but it was worth testing again to confirm.

4) Grading the same clips in ACES instead of YRGB is far easier and the final results look better than YRGB. No LUTs of any kind neeed, not even a 3D LUT. ACES is amazing and requires 1/10 the labor of YRGB.

So those are my conclusions. Other people may have different views but in the end it all comes down to personal preference anyway.
 
Hi Doug,

fair enough. Sorry if I didn't sound sure to you. Put it like this: the processing in the LUTs is a precisely, mathematically correct adjustment of the exposure shift of S-Log3 to S-Log3.

The idea is one click and you have unquestionably done the same process as happens in camera if you have an S-Log3 LUT turned on and change the CineEI ISO. It also uses exactly the same data as you would find in the ACES IDT (ACES incorporates an additional 3x3 colour space transform to and from itself, but the contrast adjustment is otherwise the same).

Hopefully that sounds less uncertain.

The thing I questioned is the lack of a simple slider in post software, which would do this even more easily and allow you to shift effortlessly between the ISOs for intended CineEI shifts, but also swiftly tweak for an iffy exposure. It just seems like a needless omission of a fundamentally useful tool that is available, basically doing the same thing, for raw.

Now, if you prefer your results by hand then great - all that matters is if you are happy with the end result. However, if you are using the primary controls to shift the CineEI, you are fundamentally not accurately shifting the ISO of log material. No point in mentioning any maths, but I would just say think slide rule.

Remember that applying the LUT is entirely non-destructive (in Resolve LUTs don't need to be clipped) so the idea would be that if you shoot CineEI on an F5, set the ISO to 800 and then load it into Resolve, one click and you are there. Alternatively, if you rated the camera that way for everything you shot, you can have the adjustment applied to everything before you start to grade (LUTCalc can also generate Resolve 1D Input LUTs for use under Project Settings). From there, you can make all the fine adjustments for shots which are a bit bright or dark, or all the other stuff that goes into grading.

As such, I'd politely say that your point 2 - 'Your LUT is still going to require some manual adjustment if the exposure wasn't perfectly at 800 EI' - is a rather unfair strawman. Of course you are likely to do manual adjustment, it's a grade!

Personally, for non-ACES I would recommend Sony's workflow of linearising your log footage with a base look after the CineEI adjustment (either one of the built-in looks or a personal one). That can get rolled into the ISO shift (so one click again) or kept separate. Again, it's non-destructive of the data but also the primary controls become the relevant tool and your initial image should look as you (and your clients and director) saw it with your MLUT on the shoot day, just ready to be polished to a shine.

That doesn't make applying a base look somehow less professional, after all it's what ACES is doing and as you say that is a brilliant workflow; it hides the heavy lifting removing the issue of subjective advice. Rest assured that I use the same underlying maths,

Ben
 
Last edited:
How can any LUT be non-destructive?

I thought it depends on 1.) the lattice quality of the LUT, 2.) perhaps more importantly how much (and how good) the interpolation of the software or hardware being used is. 3.) if the environment is floating point.

A lot of hardware manufacturers will boast the level of their interpolation for smoothing out LUTs and creating "1 billion" colours. But here we're talking about a LUT in-camera or in After Effects, Premiere, Davinci, and possibly a lot more software programs - each which is probably and likely "doing it's own thing" in processing the LUT which ultimately will affect the quality of what it does with the LUT. So to generalize and say a LUT is not destructive comes with a big caveat.

Personally I would avoid as much use of LUTs in any grading chain like the plague.

LUTs are shortcuts. You will end up fiddling with them anyways to get them work on different clips. I own impulz, koji, and a few other LUT packs... I don't end up using any of them after playing around with them.

LUTs can be destructive. You can degrade your image by using them depending on several factors.

Not everything I say is true, not everything I say is false. I did grade an entire film using Sony's F65 beta slog2/sgamut LUT which screened in theatres across Canada and is screening in USA next month at a film festival, and was picked up by a distributor and will now be playing on in-flight and bus coach entertainment systems. Ok see, no one is complaining that I used a LUT and a beta LUT at that. :happy:

But in the end more from a fundamentals point of view, I think it's better to learn how to grade from scratch because as I've said many times before one problem with LUTs is they won't always work applied to every clip you might have shot in a day (or a week) and need modification anyway so why bother if you're going to spend all that time fixing a LUT then or finding a substitute? For me it sounds as frustrating as trying out 1,000,000 combination of magic bullet look settings.

An exception to my "rules" would be with film emulation LUTs like kodak, fuji, and of course LC709 based LUTs because they would be very difficult and time consuming to try and create from scratch by hand. Well at least for me and most "average" users.

I realize I'm talking a lot about LUTs for "looks" here and I understand the desire to take shortcuts but speaking to offset LUTs specifically, I think there's even less reason to use them in post again because I don't know what it's doing to my image and it's really so easy to just do any exposure compensations yourself.

I also didn't think Resolve's "offset" wheel was the same as ISO slider but I was proven wrong recently. Slide it, and there is your ISO adjust. I also agree with Doug, and stay away from log wheels. Primary wheels do everything you need for Sony slog. If only I knew that sooner.

I guess to LUT or not to LUT is the big question. It's a hot topic with so many professional cameras and now consumer level DSLR's capable of shooting in log. I don't expect use of LUTs to go away anytime soon.
 
Last edited:
Hi Dennis,

How can any LUT be non-destructive
Click to expand...

Good question, and the answer is it depends. Taking Resolve as an example - since that is what has cropped up - the internal processing is done on floating-point values and as such references to 10-bit integers is for the benefit of the user rather than anything to do with the ongoing accuracy of operations. LUTs - cube LUTs, iluts and oluts in particular contain floating-point values which are the values to which input values (scaled from the limited 8-bit, 10-bit, 16bit etc integers) are mapped.

In the case of a 1D LUT such as the ones I posted, there are 4096 sets of output values. Since this is intended to be used as the first link in the chain, this clearly means that there are more than enough values to cover a 10-bit recording without any interpolation being involved. Furthermore, since output values in Resolve need not be clipped (check out the built-in LUTs, particularly the linearisation or base Rec709 ones in Resolve to see for yourself), you can push the ISO in an adjustment LUT to values effectively way beyond 109% IRE without losing any data. In exactly the same way ACES converts the log encoding to floating-point linear internally, for a working dynamic range where it is effectively impossible to destructively clip your image data.

Now, if you end up with floating point values which are outside of the -7% to +109% range and then feed that to another LUT, then generally you will be outside the input range and things will get destructively clipped. That is why I suggest rolling exposure adjustment and colour looks into one LUT as something like LUTCalc (or ACES) goes back to the base equation of the well defined log curve so can extrapolate as well as interpolate.

Alternatively, the cube format in particular allows for defining an arbitrary input range. Some post software throws a wobbly about this, and Resolve and Adobe have two incompatible ways of defining this, but in a Resolve cube you can say that the input range is actually 0-1.0 (default) or -25-+27 or whatever you wish. If you look at Resolve's built-in linear to log corrections, they make use of this (linear or Rec709 within 0%-100% IRE only cover around 2 1/3 stops of headroom, the input scaling is used to give values for the six, seven or eight stops that a log curve might allow in 0%-100%. In the case of a linear to anything LUT such as that, the spread of values is such that a larger mesh is tangibly better and Resolve allows for 14-bit (16384 step) 1D LUTs. I've just added such to LUTCalc testing. That's a fairly technical usage case though; most people will be working the other way - log to linear (or conventional gamma).

Back to non-destructive. So far nothing has needed to be clipped. Numbers have been shuffled around, tones changed, but distinct values are still distinct. Next step is that tone curves are generally monotonic; as the brightness hitting the sensor increases, so does the curve value (unless its a negative, then it's the other way around ;-) ). Still nothing lost.

Finally you reach white clip. A curve can reach a plateau, either as a hard clip or gradually shifting towards it, in which case any values above that are unchanged. LUTs generated in Resolve clip at an output value of 1.0 and as such, yes, anything above that will be destructively clipped, but if you look at curves such as Rec709(800%), LC709 and LC709A the output slope is not generally horizontal at the top of the input range, ie every input combination can potentially have a discrete, unique output combination and again, no image data is destroyed. Shifting from a wide gamut to a narrower one such as Rec709 can cause destructive clipping, but on the whole the floating-point nature of the processing keeps a decent amount of wiggle room.

You mentioned the interpolation techniques used, and I have to make clear that whatever artifacts that may produce is not the same as saying that the LUT is not non-destuctive. It might do things you don't like, but something that was a bit brighter than another thing will still be a bit brighter than another thing as far as Resolve is concerned.

In the 1D domain, the most common forms of interpolation are linear in nature. The thing is, a 1D mesh is typically larger than the input steps and certainly way beyond the visual acuity of human vision, so even if the material has been prodded around before applying the LUT, there should be no visible artifacts as a consequence of a well-constructed 1D LUT.

3D LUTs are somewhat different. Typically the interpolation is an extension into 3D of linear techniques, typically trilinear or tetrahedral interpolation. A smoother and arguably cleaner alternative is tricubic interpolation, but this is significantly more processor-intensive. The memory demands of having an output triplet for every possible input triplet would be huge, so mesh sizes tend to be relatively small sounding, 17x17x17, 33x33x33 or 65x65x65 being typical and then using interpolation.

Going back to the fact that tone curves tend to be monotonic, unless you are doing aggressive highly specific colour adjustments the transition between mesh points tends to be gentle and smooth - particularly from 33x33x33 over current dynamic ranges - so interpolation techniques such as tetrahedral or simplex work well. For well-formed 3D LUTs, artifacts such as banding will tend to come from other parts of the processing chain, in particular the 'dec' part of the codec.

As with 1D, interpolation technique aside in Resolve no image data need be irreversibly destroyed, just shuffled around.

Other pieces of post software may not do their processing in quite such a rigourous way, but then that doesn't auger well for any of their processing tools!

1D and 3D LUTs are used all over the shop without you being aware of them. I would be surprised if the camera doesn't use a LUT to convert linear data to the log recording in the first place; it is a big porecessing saving in a bit of kit that wants to make the post of the battery power available.

Certainly the Amira, which seems to be something of a benchmark, uses 33x33x33 LUTs across the board, either as an MLUT for LogC recording or if you shoot Rec709 directly any colour adjustments you make are converted into a 3D LUT internally to again keep the processing overhead down. I don't hear any issues there!

What I have come across is 3D LUTs where linear interpolation has been used to upsample the mesh size and then further LUTs have been applied in series. That can definitely cause artifacts and I've heard of people having issues with some commercial LUTs where I'm fairly sure that's happening. To avoid or at least mitigate this, I have always used tricubic interpolation in LUTCalc.

Clearly LUTs can potentially be destructive, particularly either through clipping in the LUT itself or through the post software throwing away data outside of -7% to 109% IRE, but they fundamentally do not need to be and certainly not in DaVinci Resolve.

Sorry for the longwinded answer, but you did ask! :-)

Ben

PS I'm not saying the offset wheel is an ISO slider - if you look at the shape of a log curve against stop, above 18% gray they tend towards a straight line (though don't quite get there). That equates to a fairly consistent offset. Below 18% gray, the divergence from a straight line becomes more and more prominent, ie the offset shrinks. Mathematically it is a doddle to do a shift correctly - it should be a trivial thing, but unquestionably, factually, playing with gains is an approximation. And, to me, a pointless bit of effort. I promise you, a 1D exposure shift input LUT has no downside at all, and does no damage in something like Resolve.

Ben
 
Last edited:
I've said many times before one problem with LUTs is they won't always work applied to every clip you might have shot in a day (or a week) and need modification anyway so why bother if you're going to spend all that time fixing a LUT then or finding a substitute?
Click to expand...

the whole point is not apply LUT - bosh - finished move on, though of course you can do if you want; Sony's looks as an example are intended as a consistent base. Look at them on set (surely the client isn't looking at uncorrected log) then apply them to your footage in post and you should be looking at what everyone saw on the monitor on shoot day. Then do what you need to to polish individual setups (or have a small toolkit of base looks if you wish),

Ben
 
Hi Ben,

As always great discussion all the way around and good insights.

With these types of discussions it's the only way closely in sorting through the mud.

I know I also mixed in the notion of 3D LUTs for "looks" versus 1D exposure shifting so it's great that you went into the distinction.

All of this just makes me wonder where people in the "industry" learn this stuff.

:shocked:
 
starcentral said:
I know I also mixed in the notion of 3D LUTs for "looks" versus 1D exposure shifting so it's great that you went into the distinction.

All of this just makes me wonder where people in the "industry" learn this stuff.

:shocked:
Click to expand...

My impression is that cleverer people than us have generally made it up as they find a need, which is why there are so many LUT formats (and differences within formats such as cubes) and why there are so many different approaches to post software and post tools.

I'm happy to just take pictures, hand over the rushes confident that I know something nice can be done with them and that my hobby / obsession is a useful background when I know it's going to be handled by a post-boffin - or by the boffin's intern's cousin for that matter! :-)
 
Ben and Dennis,

Thanks for all the great information and I see lots of solid advice that I agree with in posts from both of you. But the conversation has gotten more technical than I care about so I'm going to step away, but I just wanted to let you know I appreciate the answer to my question and read all of it. For me ACES is fantastic and makes much of the discussion about LUTs moot for me anyway.

@ Morgan
I graded a bunch of wildlife footage last night with a .47 radius sharpening and it does look good. That will probably be my new default setting for CINE EI footage. Thanks for making me experiment.
https://vimeo.com/160234425
 
Last edited:
Hi Doug,
If you have to apply sharpening in Resolve . . . get a better lens or do a better job focusing. I never have to use the sharpen control with S-LOG or any other profile with the FS7.



And this editor I know who does this standard does need new glasses :-)

 
Last edited:
You must log in or register to reply here.
Back
Top