A7s: S-log2 too noisy? Then underexpose cine4
- Thread starter macgregor
- Start date
With "flatter" I meant that cine2 raises the blacks even more than cine4, so I was wondering if it wouldn't be better to use cine2 instead. Perhaps there is and advantage in using cine4 which I have not noticed yet? Until now in my own tests I see virtually no difference between cine2/cine4 with raised blacks or log2 in terms of dynamic range. They all seem to capture all the DR the sensor is able to deliver.
Another interesting thing: highlights seem to peak at the same point using an ISO of 640 or 800 with the cinema gammas and when using Slog2 at ISO 3200. I have come to believe that the "native" ISO of the camera is around 640 or 800. When using lower ISO values I see some artifacts in the blacks, which are an indication of negative gain. The high ISO value of Slog2 is probably to make sure that the automatic exposure is correct.
Another interesting thing: highlights seem to peak at the same point using an ISO of 640 or 800 with the cinema gammas and when using Slog2 at ISO 3200. I have come to believe that the "native" ISO of the camera is around 640 or 800. When using lower ISO values I see some artifacts in the blacks, which are an indication of negative gain. The high ISO value of Slog2 is probably to make sure that the automatic exposure is correct.
starcentral
Veteran
I for one would love for someone to show me aside from a step chart where an image is going to benefit from the extra dynamic range offered by slog2 over the cinegammas.
Samuel H
Major Contributor
I think that if you have to shoot a subject in the shade with a sunny background, slog2 can keep the highlights alive and still not be too noisy in the subject, with cinegammas you'll have to decide to either clip something in the highlights or have a very noisy subject with very little gradation.
I'm finishing the search for my preferred settings, it should be done this afternoon. And I just received my proND set. So I will probably go out to test them both soon, I'll make sure to include such test vs the cinegammas.
I'm finishing the search for my preferred settings, it should be done this afternoon. And I just received my proND set. So I will probably go out to test them both soon, I'll make sure to include such test vs the cinegammas.
paulcurtis
Veteran
Hasn't Black Level been done to death on the FS100 and FS700 over the past few years? Increasing your black level is just raising where the blacks are recorded, basically loosing space in the movie container file. It doesn't change what is recorded, just how it is stored. Black gamma again doesn't *add* any tonal detail, it just bumps part of the curve up at the bottom.
A log curve is a very special curve that can be easily converted back into high dynamic range in post (i.e. the full range stored)
Any attempt to make a log looking curve is kinda missing the point because you have no clear way to bring the full range back in a quantifiable way. Sure, if you want to raise your blacks because you think it looks loggy, then in post bring them down again to make them black, that's your choice. But you're just loosing range that way.
The point Samuel makes about 'does Slog2 produce less stored values because it is squeezed in video levels' is a valid question, especially if cine4 doesn't. I think the answer is no, because the file is not stored in RGB 8 bit levels but YUV.
Funnily enough i have a bunch of night car driving stuff to do shortly so i'm testing the A7s and seeing whether i can use that as it would be lovely and lightweight (but rolling shutter may put an end to that). So i am about to start looking in detail at these kinds of results.
As for the original post, yes cine4 and Slog2 could share the same dynamic range (as it does on the FS700) but it's really the curve between those two extreme points that matters.
Also cine4 may suffer colour clipping before Slog2 does too, i suspect that could be a big issue.
cheers
Paul
A log curve is a very special curve that can be easily converted back into high dynamic range in post (i.e. the full range stored)
Any attempt to make a log looking curve is kinda missing the point because you have no clear way to bring the full range back in a quantifiable way. Sure, if you want to raise your blacks because you think it looks loggy, then in post bring them down again to make them black, that's your choice. But you're just loosing range that way.
The point Samuel makes about 'does Slog2 produce less stored values because it is squeezed in video levels' is a valid question, especially if cine4 doesn't. I think the answer is no, because the file is not stored in RGB 8 bit levels but YUV.
Funnily enough i have a bunch of night car driving stuff to do shortly so i'm testing the A7s and seeing whether i can use that as it would be lovely and lightweight (but rolling shutter may put an end to that). So i am about to start looking in detail at these kinds of results.
As for the original post, yes cine4 and Slog2 could share the same dynamic range (as it does on the FS700) but it's really the curve between those two extreme points that matters.
Also cine4 may suffer colour clipping before Slog2 does too, i suspect that could be a big issue.
cheers
Paul
Samuel H
Major Contributor
So, I have my Flaat settings dialed down. It's basically slog2 with cinema color, a ton of saturation, and some minor color tweaks (I will present it to the public when it is thoroughly tested).
And having done that, I went out to do this slog2 vs cine4 test.
This is the usual scenario where I'm crying for DR: nice blue sky, with beautiful clouds, but my subject is in the shadow, and I can't bring in any lights.
Light meter readings:
* f/11.0-2 in sunlight
* f/2.8-5 in the shadow, in the far side of the street
So, the shadow in the far side of the street is 3.8 stops darker than sunlit areas.
I exposed just so the clouds would look nice: there can be some clipping, but it has to look like a blue sky with beautiful clouds. That meant:
* for slog2, f/8.0, 1/50s, ISO 3200, 6 stops of ND
* for cine4, f/3.5, 1/50s, ISO 200, 6 stops of ND
I applied 20-235 output levels to the cine4 image, to get back superwhites and match the black point of slog2.
And this is the resulting image with both profiles (open in new tab to see it full res):
slog2:
cine4:
It's not just that the slog2 one is brighter in the shadows. You can see detail in the door, in the trees, in clothing, etc, that was basically lost with cine4. It's just one stop, but you sometimes need that one stop.
And having done that, I went out to do this slog2 vs cine4 test.
This is the usual scenario where I'm crying for DR: nice blue sky, with beautiful clouds, but my subject is in the shadow, and I can't bring in any lights.
Light meter readings:
* f/11.0-2 in sunlight
* f/2.8-5 in the shadow, in the far side of the street
So, the shadow in the far side of the street is 3.8 stops darker than sunlit areas.
I exposed just so the clouds would look nice: there can be some clipping, but it has to look like a blue sky with beautiful clouds. That meant:
* for slog2, f/8.0, 1/50s, ISO 3200, 6 stops of ND
* for cine4, f/3.5, 1/50s, ISO 200, 6 stops of ND
I applied 20-235 output levels to the cine4 image, to get back superwhites and match the black point of slog2.
And this is the resulting image with both profiles (open in new tab to see it full res):
slog2:
cine4:
It's not just that the slog2 one is brighter in the shadows. You can see detail in the door, in the trees, in clothing, etc, that was basically lost with cine4. It's just one stop, but you sometimes need that one stop.
High contrast outlining AKA "highlight aliasing" rears its ugly head. Sadly it isn't just a fully clipped background that has caused it. Anyway, as far as usable range, slog 2 "has it". The curve just cant be replicated with other settings. I think one stop difference is pretty accurate.
gaffney film
Active member
Where do you see that?
Samuel H
Major Contributor
Left sky-building contour.
BTW: it looks worse on slog2 than on cine4 here, but I think it's just because cine4 is f/3.5 and I was focusing at the people, so that wall is slightly oof.
I didn't test all the other options, but I believe in this case only slog2 has the DR I need to have the people in the shade decently exposed and the sky non-clipped. You can use out-of-the-box slog2 (PP7), but other gamma curves won't do.
paulcurtis
Veteran
Okay, i've done some initial tests. It's my belief that processed through a proper log pipeline the ISO 3200 is not more noisy that Cine4.
I shot a transmission wedge, i set zebras to 100 and tried to get the same wedge to overexpose. No black gamma and knee manual to 100 no slope.
I set the same f stop and shutter speed and tried to just use the ISO to match between Slog2 and Cine4. I found that ISO 3200 matched ISO 640 on those two fairly well. The A7 is very difficult to monitor properly and judge exposure. I will do these again with an O7Q so i can use the waveforms properly.
Take the XAVC into Resolve (transcoding changed the image and AE/PPro are unclear with what they are doing). I also say that i don't trust Resolve either but right now this is the only choice.
Take the two clips in but on the Slog2 add the Sony Slog2 to Rec709 gamma curve, a 1D LUT, it's just gamma remember. I gained the waveform so that both Slog2 and Cine4 met the same point and white balanced the image with gain/gamma. This throws up an interesting observation. The white balance of the camera isn't that accurate (i was shooting 56K light with 56k WB but it wasn't right). To WB log is fairly easy because the balance between the channels seemed quite linear but the cine4 was more difficult because the WB was very non linear across the range.
The end result is that from a range viewpoint i'd say Slog2 just about has the edge and from a noise perspective i think that Slog2 is actually a little less noisy in this particular case.
The key is that i am using a 1D LUT to turn Slog2 into a display gamma, i am not *grading* from a log signal with raised blacks and gammas.
I'm not saying that anything that has been done before is wrong and this is just a first test, there are many steps that mistakes can be made. Also i am not able to tell whether the Slog2 signal has less steps than cine4. But the log format is a easier place to grade from IMHO, it fits into pipelines better than cine4. You can see the way the white rolls off on the cine4 making it look brighter whereas the log wedges are more uniform. You have to grade log properly whereas cine4 works better out of the camera.
Also i've not messed with knee and black gamma to change the curve i just wanted to compare base SLog2 ISO with it's counterpart in Cine4. Of course Cine4 can go lower than 640, that could be very useful too.
cheers
Paul
I shot a transmission wedge, i set zebras to 100 and tried to get the same wedge to overexpose. No black gamma and knee manual to 100 no slope.
I set the same f stop and shutter speed and tried to just use the ISO to match between Slog2 and Cine4. I found that ISO 3200 matched ISO 640 on those two fairly well. The A7 is very difficult to monitor properly and judge exposure. I will do these again with an O7Q so i can use the waveforms properly.
Take the XAVC into Resolve (transcoding changed the image and AE/PPro are unclear with what they are doing). I also say that i don't trust Resolve either but right now this is the only choice.
Take the two clips in but on the Slog2 add the Sony Slog2 to Rec709 gamma curve, a 1D LUT, it's just gamma remember. I gained the waveform so that both Slog2 and Cine4 met the same point and white balanced the image with gain/gamma. This throws up an interesting observation. The white balance of the camera isn't that accurate (i was shooting 56K light with 56k WB but it wasn't right). To WB log is fairly easy because the balance between the channels seemed quite linear but the cine4 was more difficult because the WB was very non linear across the range.
The end result is that from a range viewpoint i'd say Slog2 just about has the edge and from a noise perspective i think that Slog2 is actually a little less noisy in this particular case.
The key is that i am using a 1D LUT to turn Slog2 into a display gamma, i am not *grading* from a log signal with raised blacks and gammas.
I'm not saying that anything that has been done before is wrong and this is just a first test, there are many steps that mistakes can be made. Also i am not able to tell whether the Slog2 signal has less steps than cine4. But the log format is a easier place to grade from IMHO, it fits into pipelines better than cine4. You can see the way the white rolls off on the cine4 making it look brighter whereas the log wedges are more uniform. You have to grade log properly whereas cine4 works better out of the camera.
Also i've not messed with knee and black gamma to change the curve i just wanted to compare base SLog2 ISO with it's counterpart in Cine4. Of course Cine4 can go lower than 640, that could be very useful too.
cheers
Paul
"Cine4 can go lower than 640" Yes, but then you get MORE noise not less. The cam seems to apply negative gain at lower ISO settings. 640 seems to be the native ISO of the camera. It matches Slog2 at 3200, as you said, which explains why Slog2 has a minimum ISO of 3200 - anything less would deteriorate the image.
nyvz
Veteran
Matched how? White clip level? Black level? Middle grey? I found ISO1000 CINE4 to match SLOG2 ISO3200 for white clip. I'm seeing a lot more noise in your CINE4 step chart than the SLOG2. This makes sense since that means at ISO640 you are underexposing your cine4 by 2/3 stop and that would explain why it looks noisier than the SLOG2 when you grade it to match SLOG2 ISO3200. Also a consideration is that the camera employs a different form of gain at a certain point that causes it to perform better at certain gain settings than certain low gain settings (http://legault.perso.sfr.fr/sony_a7s_measures.html). Unfortunately Sony doesn't tell us what gain level or mode is applied and the ISO level is a combination of the hardware gain and the gamma curve so we are left to guess what one ISO means in one gamma curve compared to another gamma curve. One thing you do see from that read noise graph is that there is about a 1-1/3 stop jump in DR when the camera switches gain modes and the lowest new gain has similar DR to a low gain mode with an ISO about 2-stops lower. Unfortunately we have to test to figure out where that jump happens for any particular PP/gamma.
My guess would be that gain applied at SLOG2 ISO3200 is the same as ISO1000 CINE4 and then furthermore from those read noise tests that there may be similar noise/DR at ISO1000 CINE4 as there is at ISO250, and that ISOs in between may be inferior.
paulcurtis
Veteran
You can only really match with white clipping point without some decent exposure tools. So i did 100% zebra and then matched the wedge at which that clipped (which is where the arrow was). Of course the difference between 640 and 1000 is quite a lot but actually quite difficult to see on the A7s. So you might be right in that the ISOs are closer then they seem - i need to redo this with the O7Q to get some accurate exposure
The point at which the gain gets better is 3200, there's a double exposure bump with the A7s, there's a really tech blog somewhere with a bunch of experiments and analysis over this.
It's also possible that 100% isn't ideal either, because 100% isn't clipping for either Cine4 or Slog2, if i choose 105% then it's not a single point but a range and that's difficult to tell too. I choose 100% at the mark with the idea that the wedges after can handle superbrights but the signal is doing something odd there (some kind of knee in play)
I think the point i'm making though is that if you apply the proper 1D Slog2 LUT to the footage and work from there the noise isn't as bad. The issue is really that if someone is grading from log then because of the raised black point is everyone doing the right thing? Taking the black level down and working from there is not quite the same operation as the LUT is doing. The pivot points are different so it's not so easy to get blacks to where they should be.
Anyway, i shall repeat this with the O7Q monitoring so i can hit a few different ways of matching - mid grey for example.
cheers
Paul
paulcurtis
Veteran
I've been thinking about this, i'm wondering now:
What is 3200 Slog if actually a digitally gained version of ISO640 purely to be able to create the log curve. When a suitable log LUT is applied this digital gain is reversed so it's back to 640. So 640 and 3200 are like for like in terms of noise floor.
On the FS700 it's very similar, you have to gain a RAW signal in post by 4 stops to get it to match Slog (once you remove resolve from the equation which automagically does stuff)
to complicate matter it's might *not* be a straight gain because it may not gain the highlights so much, the reason is to compensate for the linear encoding of the shadows (i wrote a blog post about this on our blog)
So in theory if i can accurately match (instead of the ball park match off the LCD) Slog2 and Cine4 the best i can (the gamma curves are different so they're never really going to match - ideally i'd do it around mid grey). Then side by side different ISO combinations maybe we can confirm that Slog2 is actually the base ISO of the camera digitally gained.
They couldn't call the base 640 because of the image the Slog2 produces so it's an 'effective' 3200 but the camera is not running at 3200.
The reason i suggest this: because i think the noise is quite similar, there's more of a gap between 640 and 3200 on the stills side. (Of course it could be the other way around and 640 is digitally generated from a real base of 3200).
Does this sound plausible?
cheers
Paul
Ben Canales
New member
@Macgregor- this setting sounds amazing. Question- what picture profile are you using for this setting?
I think I'm misunderstanding something basic- but since Cine4 is available in most of the Picture Profiles (PP1 - PP7) which PP setting are you using for these parameters?
Thanks for posting this information!
I think I'm misunderstanding something basic- but since Cine4 is available in most of the Picture Profiles (PP1 - PP7) which PP setting are you using for these parameters?
Thanks for posting this information!
macgregor
Veteran
Hey, is there an article, post or video about this? Interested to see how noise / DR changes on cine 4 on different ISOs.
macgregor
Veteran
PP are just memory slots. You can change the settings to whatever gamma you want. People keep saying Slog2 is in PP7 but for example back on the day I just created an Slog profile on PP1.