A7s Exposing up and bring down examples

paulcurtis

Veteran
I thought i'd start a thread for this example.

A difficult Slog2 scene, dusk, dark and detail in the wrong part of the Slog curve. But if you expose higher up then bring the exposure down in post you can maximise the noise performance. You are basically robbing the highlights and putting the data in the shadows.

This is the same scene, shot 3200, 6400 and 12800. In SpeedGrade the exposure is balanced to match(ish). The Slog2 curve was enough to retain the highlights even at 12800 ISO.

Because the Slog2 is scene linear it means that a post adjustment like this is photometrically accurate, something you cannot really do with the other profiles.

The LUT on top was the sony provided Slog2/SGamut -> Rec709

You can see that exposing 2 stops up and then bringing that down in post has improved the shadow noise handling a lot. Compare the 3200 vs the 12800. It also means we are avoiding the very bottom end which goes purple. The matrix has a levels adjustment as well so that the shadows are more visible still. So you may think that 12800 is more noisy but it isn't really when you look at it this way and in the shadows.

I've linked this image to my host because DVX is scaling the images and recompressing which makes it all a bit pointless really!

I'm not convinced the Sony provided LUTs are ideal for this camera though, but if you're going to be doing exposure adjustments you really need a way to turn Slog2 into linear space (or log if that's what your grading app is doing) then you can apply a gain (for linear space) or an offset (for log space) to change the exposure.

hth
Paul

iso.jpg
 
Very Interesting.
Just for clarification are you keeping the aperture and shutter settings constant between these examples? I'm sure many of us know that T-stop and exposure time are the only things that change actual exposure. The ISO just changes how the sensor deals with the exposure.
I think I'm right in assuming that what you are doing here is increasing the analogue gain and then doing a digital attenuation in post? Or are you decreasing the actual exposure in the higher ISO shots?
How would this compare to physically exposing higher and then doing a digital attenuation?
 
Not only is there less noise, there's more life in the high ISO shots.

It's the other way round for me. Less noise in the lower ISO, at least where it matters most - in the mids. The shadows might be fine at the higher ISO, but the mid tone is a mess. Skin lives in the mids...


EDIT: Ignore this, I can't read...
 
Last edited:
Those splotchy 3200ISO and 6400ISO look like when I used to shoot on a GH2 and thought I could bring back some shadow information. Yuck! Back to workarounds on "how to make your videos less videoish."
 
Very Interesting.
Just for clarification are you keeping the aperture and shutter settings constant between these examples? I'm sure many of us know that T-stop and exposure time are the only things that change actual exposure. The ISO just changes how the sensor deals with the exposure.
I think I'm right in assuming that what you are doing here is increasing the analogue gain and then doing a digital attenuation in post? Or are you decreasing the actual exposure in the higher ISO shots?
How would this compare to physically exposing higher and then doing a digital attenuation?

In this case it was with ISO only 3200-6400-12800, the whole point of the sensor is the large pixels meaning that analogue gain is a valid way of changing exposure.

The technique gives the results shown, no matter how you look at it the end result of using a higher exposure and pulling that down results in an image with visibly less noise.

The choice sony made to rate the base ISO of the Slog2 at 3200 is really about pushing the noise floor up and striking a balance for a wide range of shooting situations. This is no different to rating the camera at 1600 instead to gain a stop on the bottom.

cheers
Paul
 
It's the other way round for me. Less noise in the lower ISO, at least where it matters most - in the mids. The shadows might be fine at the higher ISO, but the mid tone is a mess. Skin lives in the mids...

Ah, but less effective noise in the final image by doing this.

There are two different things at play here. One is that no matter what the ISO the very bottom of the range of the camera is going to be all noise. There is no black, just the point at which you say that's all noise now. The other is the general noise across the image from analogue gain.

On this camera the analogue gain is remarkable similar from 3200 all the way to 12800 and beyond in some cases. There isn't an appreciable different between those really.

So use the pretty clean ISO noise at a higher analogue gain to help create a lower noise floor.

That's what the images show, the general patina of purple and mush in the 3200ISO is made much cleaner by exposing higher and pulling that image down to the same levels.

Hope this makes sense?

cheers
Paul
 
Those splotchy 3200ISO and 6400ISO look like when I used to shoot on a GH2 and thought I could bring back some shadow information. Yuck! Back to workarounds on "how to make your videos less videoish."

That's pretty unfair because i am testing worst case scenario and i'm levelling up the result. The GH stuff looks like this normally! (flame suit on).

This is a compressed codec too

And have you ever raised the blacks on a film scan? :)

cheers
Paul
 
Makes complete sense to me, you probably arrived at it sooner n'I did though! Took me a bit of time to understand it. I also agree that there aren't any good workflows for the actual camera at this moment... time to make stuff up. haha.

Guess it's not a bad thing, either, because it sort of frees you up to get away from the stock settings.

Me feels the camera's doing a lot of fair... now if it were only like 800 cheaper!
 
I've found most stock SLOG2 LUTs to be awful and are poisoning the well when it comes to people's opinion over what is possible with the camera.
 
Those splotchy 3200ISO and 6400ISO look like when I used to shoot on a GH2 and thought I could bring back some shadow information. Yuck! Back to workarounds on "how to make your videos less videoish."

They don't look like that when you don't abuse them. I have no idea what people are doing to their footage.....it's like looking at a different camera. When you step away from scope chimping and tests, you'll be more likely to just SEE the image and get it to a place that looks wonderful to you. That's my take. More creating, less pixel peeping!
 
That's pretty unfair because i am testing worst case scenario and i'm levelling up the result. The GH stuff looks like this normally! (flame suit on).

This is a compressed codec too

And have you ever raised the blacks on a film scan? :)

cheers
Paul

I understand what you are saying and I thank you for doing these tests including Kholi and others. It just seems like the more I see of this camera, the less enthusiastic I get. I'm sure people will learn from your findings and make some damn great videos up ahead. Wish someone could just take one day and shoot a project with it. It's long overdo!

Also is there anywhere I could download some actual XVAC somewhere or could someone make one downloadable? Would really appreciate it!
 
I understand what you are saying and I thank you for doing these tests including Kholi and others. It just seems like the more I see of this camera, the less enthusiastic I get. I'm sure people will learn from your findings and make some damn great videos up ahead. Wish someone could just take one day and shoot a project with it. It's long overdo!

Also is there anywhere I could download some actual XVAC somewhere or could someone make one downloadable? Would really appreciate it!

It certainly wasn't like the first days of the 2.5K BMD camera, when that footage from JB trickled out it consistently hit home that the camera's image was not to be questioned haha. I was sold so hardcore on it instantly, and then subsequent releases of the other cameras.

But, now that I've actually had the A7S in hands I think it's going to get better as more people explore. It just isn't as immediate. Digging into menus... I haven't done that since BMD cameras hit X_X.

On testing etc.. I have a major heart for guys like Paul and Sam H, cpc, demistate, all of the guys jumping on scopes while I'm out doing the "how does this feel" half of exploring. We all are arriving at similar conclusions, and everyone else can pick up the info that they need. Personally, even though I'm nowhere near as technical as those guys, I absolutely have to know what I can hand off to someone to cut, color, or deliver and that means sucking up knowledge from these other guys while peeping the crap out of footage.

I do a lot of personal shooting, when it comes to getting paid more than not I have to hand over drives. Doesn't matter if it's the crash cam or not, you just need to know that you're able to get the aesthetic and technical bests out of a camera if you're suggesting it to someone, recommending it, etc. Like Pauls' mentioned, I can't go out and say "Yeah this was shot SLog2" to someone, their colorist gets it and tries to apply a LUT and has no idea what the heck is going on.

This is bad news.

THere's room for it allll, at least for me!
 
Ah, but less effective noise in the final image by doing this.

There are two different things at play here. One is that no matter what the ISO the very bottom of the range of the camera is going to be all noise. There is no black, just the point at which you say that's all noise now. The other is the general noise across the image from analogue gain.

On this camera the analogue gain is remarkable similar from 3200 all the way to 12800 and beyond in some cases. There isn't an appreciable different between those really.

So use the pretty clean ISO noise at a higher analogue gain to help create a lower noise floor.

That's what the images show, the general patina of purple and mush in the 3200ISO is made much cleaner by exposing higher and pulling that image down to the same levels.

Hope this makes sense?

cheers
Paul

Hang on, I'm being dopey. I thought the bottom left image said 1280, not 12800! Yes, it makes sense to me, that's how I do it in Clog,but you have to be careful with skin. Overexposed log skin can look alien.
 
Sadly i'm tied to my desk rendering on many workstations and nuking away so the only footage i can grab is out my window and look at the technical stuff. We shot last weekend, part of much larger project, but that was full on FS700/O7Q/4K RAW with a quick bit of A7s just because i could.

Popped a 12mm on this full frame with a helicoid adapter and focused 3cm away from an actors face for a distorted 'going ape moment', wonderful result that i think only that combination could have gotten. That's what the A7s is for me, it can get into places that others cannot both physically but also the full frame 'look'.

A not really graded grab from it, there's nothing technical here, it's a heat of the moment frame that serves the story very well, dvx has compressed it to nothing, yellow splotches and all (well okay, i'll have to get rid of them... :)

12mm.jpg

This weekend will be a proper break, so i'll take my new camera and just enjoy.

Yes, i am looking for problems. When on location, under pressure i need to understand where the tools break and how to use them the best. When there are many folk contributing you don't want to be faffing around experimenting. Hitting scopes will often tell you more about the image than a subjective view. Look at the gaps in the histogram with speed grade to see where the camera can fall down. Also i have to get the camera playing well and matching other cameras - so the slog route is key.

Big shout out to Kholi too, who is way more knowledgable than his modesty suggests.

cheers
Paul
 
Last edited:
Hang on, I'm being dopey. I thought the bottom left image said 1280, not 12800! Yes, it makes sense to me, that's how I do it in Clog,but you have to be careful with skin. Overexposed log skin can look alien.

Being dopey is usually my trick!

yes, you're right although the beauty of slog is that it doesn't suffer the knee that other gamma curves tend to have, which is why i'm happy moving down a couple of stops

cheers
Paul
 
Back
Top