The way Cinema5D measures DR is to count down the number of stops from the clipping point to the point in the shadows where noise = SNR 2. To say it has 1.5 stops more in the highlights is as meaningless as to say it has cleaner shadows because all that is changed is the exposure indices (EI). So if someone wanted it clarified exactly what constituted the highlights, it's the number of stops above middle gray. Using the Pocket 6K as an example it has only two actual base ISO's, low and high, but on Low there are 3.9 stops of highlight latitude at EI 100 versus 7.3 stops of highlight latitude at EI 1000. Nothing has changed sensor wise except the mapping of middle gray. And it is this way for Arri and also Sony.
RED Komodo to start limited shipping soon for $6k
- Thread starter ahalpert
- Start date
The way Cinema5D measures DR is to count down the number of stops from the clipping point to the point in the shadows where noise = SNR 2. To say it has 1.5 stops more in the highlights is as meaningless as to say it has cleaner shadows because all that is changed is the exposure indices (EI). So if someone wanted it clarified exactly what constituted the highlights, it's the number of stops above middle gray. Using the Pocket 6K as an example it has only two actual base ISO's, low and high, but on Low there are 3.9 stops of highlight latitude at EI 100 versus 7.3 stops of highlight latitude at EI 1000. Nothing has changed sensor wise except the mapping of middle gray. And it is this way for Arri and also Sony.
G
Guest
Guest
Haha! So organic!
Ryan Farnes
Well-known member
I hear ya.
In 2020, it is simply odd to see a major camera company behave like a kickstarter campaign with updates and apologies for delays spread across 3-4 websites on the internet.
ahalpert
Major Contributor
Cinema5D did not use the imatest for this comparison. They shot a lit scene with both cameras exposed normally at ISO 800. Then they gradually opened the aperture until each camera clipped. So I think it's fair to say that at ISO 800, the Alexa has 1.5 stops more headroom than the Helium.
https://nofilmschool.com/2017/02/shootout-arri-alexa-mini-vs-red-epic-w-helium
Last edited:
Eric Coughlin
Veteran
And they downvoted your post too. There has been a recent anonymous group of "DVXuser Downvoter Fappers (DDF)" who can no longer get off to hardcore pornography so have resorted to downvoting as their only means of getting off, as they enjoy the sadistic nature of putting others down online while they get up. It is really best to just understand what a downvote implies the downvoter is doing at the time and not criticize, but rather, to feel pity.
Last edited:
I agree it's more headroom because the Arri exposure indices puts middle gray 1.5 stops down the IRE scale for the same EI, which is consistent with the second observation that the Komodo was cleaner in the shadows. But if you chose EI 400 on the Arri and exposed for the scene, you'd lose a stop of that headroom but have less noise in the shadows because now you're letting more light on the sensor. The notation of 1.5 stops more headroom can be a false narrative of superior performance of the sensor when it's merely a consequence of the EI choice made, and appears so in this case. 1.5 stops of anything sounds like "big advantage" when "somewhat cleaner in the shadows" sounds miniscule. If the cameras had identical DR, the offset is the same. I'm not saying that they are.
ahalpert
Major Contributor
Sure, but you can't remap the tones at a given ISO setting. Chase both cameras down to their lowest ISO and the Alexa will still have more headroom.
Even if they had identical DR and it was simply allocated differently, the headroom would be more useful to me. The most extreme contrast situations I run into involve retaining sky detail or taking blown-out windows, plus having practicals in frame during night scenes. And psychologically I think people are used to looking at areas of darkness they can't discern but not areas of brightness since our eyes adjust for the brightest part of the scene.
I think an ISO 800 comparison is fine to make but it would have been nice to know the exact difference in the shadows. What do you think?
Even if they had identical DR and it was simply allocated differently, the headroom would be more useful to me. The most extreme contrast situations I run into involve retaining sky detail or taking blown-out windows, plus having practicals in frame during night scenes. And psychologically I think people are used to looking at areas of darkness they can't discern but not areas of brightness since our eyes adjust for the brightest part of the scene.
I think an ISO 800 comparison is fine to make but it would have been nice to know the exact difference in the shadows. What do you think?
LochnessDigital
Active member
For what it's worth, RED has since re-calibrated the ISO's up 1 stop so that DR is now allocated 1 stop better in the highlights at the expense of 1 stop of shadows. Essentially, ISO 800 became 400. 1600 became 800. So an 800 vs 800 comparison between Alexa and Helium today would only show ~1/2 stop better highlights in Alexa now.
There's also better processing happening in the latest IPP2 workflow including some highlight recovery stuff, which was not done in this test. You can see the results of the improved processing here: https://vimeo.com/217270940
So things are a bit closer these days than they were.
ahalpert
Major Contributor
I bet it will be able to do 5K 60P with minimal crop, especially since the sensor is large for S35
You remap the tones by changing ISO. I would not want to compare sports cars in just 5th gear when both cars have 8 or 9 speed speed transmissions.
https://www.arri.com/resource/blob/...483e/2017-03-alexa-logc-curve-in-vfx-data.pdf
Comparing cameras at the same ISO setting is a futile exercise because there is no standard for defining digital ISO. Different manufacturers place middle gray at different points on the DR range for the same ISO. Most video cameras define middle gray as 3 to 4 stops below clipping. Except for when gain structure is actually changed, ISO settings have no effect on how the sensor captures light with raw cameras. Where to place middle gray exposure to allocate highlight and shadow ranges is a matter of creative choice regardless of the nominal ISO setting.
Arri has the broadest first stage AD range at 22 bits over two 11 bit gain stages combined. No one else does anything similar.
All of these cameras have significantly wider DR and better sensitivity than any film stock I ever shot.
Arri has the broadest first stage AD range at 22 bits over two 11 bit gain stages combined. No one else does anything similar.
All of these cameras have significantly wider DR and better sensitivity than any film stock I ever shot.
Last edited:
ahalpert
Major Contributor
Maybe I'm missing something, but there must be a way to make an apples-to-apples comparison of 2 cameras' DR and highlight roll-off, either by comparing their signal at base ISO or by comparing where they allocate stops of DR at whatever corresponding ISOs map the same absolute brightness level to middle gray. Right?
As for film DR vs digital - I never shot on film enough to form an educated opinion. But it seems to me as a viewer that even 16mm film holds richer blue tones in the sky than well-lit Hollywood Alexa-shot movies, though I suppose there's a difference between color fidelity at the extremes of exposure and highlight clipping/shadow crushing.
As for film DR vs digital - I never shot on film enough to form an educated opinion. But it seems to me as a viewer that even 16mm film holds richer blue tones in the sky than well-lit Hollywood Alexa-shot movies, though I suppose there's a difference between color fidelity at the extremes of exposure and highlight clipping/shadow crushing.
morgan_moore
Major Contributor
Its simple you get a wide dr scene open up til the highlights just dont clip. Then you check noise in the shadows.
Simples. (yet most of the internet doesnt know how to do it)
Simples. (yet most of the internet doesnt know how to do it)
John Brawley
Well-known member
There’s no standard way to measure DR other than a “theoretical” model that works from the maths that comes from the sensor spec. This is how most manufacturers calculate DR. Dirty secret.
The reason you see so much variation when you test for DR in the real world is that no one can agree what is acceptable noise. Where you put the noise floor is what determines your DR. If you can put up with a lot of noise then you can have a lot more DR. If you hate noise then the DR is much less.
In fact there’s no standard for saying what an acceptable or not acceptable level of noise is. It’s a subjective judgement.
Anyone remember when the RED dragon was going to be 20 stops ?
http://www.reduser.net/forum/showthread.php?92158-Enter-the-Dragon
Look at the first Post, an image of a DR chart. So what happened ?
A DR chart seems like the way to test it, but there’s so many subjective variables involved.
Thus, you then get meaningless phrases like “usable DR”. I hate ever talking or even looking at the manufacturers claim of DR. Best way to test DR is to compare to a camera you know. It’s only by doing bracketed exposures that you put through your own workflow to see what you can squeeze from the sensor.
JB
The reason you see so much variation when you test for DR in the real world is that no one can agree what is acceptable noise. Where you put the noise floor is what determines your DR. If you can put up with a lot of noise then you can have a lot more DR. If you hate noise then the DR is much less.
In fact there’s no standard for saying what an acceptable or not acceptable level of noise is. It’s a subjective judgement.
Anyone remember when the RED dragon was going to be 20 stops ?
http://www.reduser.net/forum/showthread.php?92158-Enter-the-Dragon
Look at the first Post, an image of a DR chart. So what happened ?
A DR chart seems like the way to test it, but there’s so many subjective variables involved.
Thus, you then get meaningless phrases like “usable DR”. I hate ever talking or even looking at the manufacturers claim of DR. Best way to test DR is to compare to a camera you know. It’s only by doing bracketed exposures that you put through your own workflow to see what you can squeeze from the sensor.
JB
Mitch Gross
Veteran
To be fair, most manufacturers do actually perform a physical test. They just differ on how they interpret it.
You can use an IMATEST or a DSC XYLA chart. DSC actually makes three versions of the XYLA, including one that delineates 26 stops of DR. Must be a real treat to manufacture to that tolerance!
Turn on the rear-illuminated chart, point the camera at it and switch off all other lights. If you don't know the native ISO (like when you're still developing the camera), set the lens at a given aperture and then dial in the ISO (translate: adjust gains on analog output of sensor) until you get maximum range from clip point to noise floor. "Adjust gains" is misleading because you always have to apply some transfer function value to the analog signal coming off the photosites in order for it to mean something. "Gain" would just be a definition of what number you are multiplying by -- everything is a gain setting, even if that number was one.
Anyway, you take that maximum range and decide at what point the noise is too high to get a usable signal (this is where some companies like to fudge things). This is a straight line curve until you apply your Log or whatever, but it does define your Dynamic Range. Your curve will determine where middle gray lands on the range, thereby giving you your over/under.
Note that there is SO MUCH MORE to a camera than just the sensor. I have seen the same sensor get dramatically different DR based on the design of the A/D conversion board placed after it. That's a different topic, but it does explain how RED initially tested more than 20 stops out of a sensor that later yielded a camera with much lower DR. They didn't lie, but I don't think I would have shared that technical test as it would never match the eventual real world performance. I will give them the benefit of the doubt as they were still a fairly young company at the time.
You can use an IMATEST or a DSC XYLA chart. DSC actually makes three versions of the XYLA, including one that delineates 26 stops of DR. Must be a real treat to manufacture to that tolerance!
Turn on the rear-illuminated chart, point the camera at it and switch off all other lights. If you don't know the native ISO (like when you're still developing the camera), set the lens at a given aperture and then dial in the ISO (translate: adjust gains on analog output of sensor) until you get maximum range from clip point to noise floor. "Adjust gains" is misleading because you always have to apply some transfer function value to the analog signal coming off the photosites in order for it to mean something. "Gain" would just be a definition of what number you are multiplying by -- everything is a gain setting, even if that number was one.
Anyway, you take that maximum range and decide at what point the noise is too high to get a usable signal (this is where some companies like to fudge things). This is a straight line curve until you apply your Log or whatever, but it does define your Dynamic Range. Your curve will determine where middle gray lands on the range, thereby giving you your over/under.
Note that there is SO MUCH MORE to a camera than just the sensor. I have seen the same sensor get dramatically different DR based on the design of the A/D conversion board placed after it. That's a different topic, but it does explain how RED initially tested more than 20 stops out of a sensor that later yielded a camera with much lower DR. They didn't lie, but I don't think I would have shared that technical test as it would never match the eventual real world performance. I will give them the benefit of the doubt as they were still a fairly young company at the time.
morgan_moore
Major Contributor
Continuing on John theory (true) that acceptable noise drives a dr figure it is actually variable job to job as a doc might have a different level of acceptability to a more glossy commercial or product job
morgan_moore
Major Contributor
It would be easy to define noise and come up with a measurement standard - maybe 75-25 s:n ratio
Mitch Gross
Veteran
Not so easy as different sensor designs create different forms of noise. Some more distracting than others. Blue sparkle, dead greens, illegal reds, on and on. Can't just stick a number on it and walk away.
ahalpert
Major Contributor
Lol re:manufacturing tolerance
Re:noise - I was reading about the different types of sensor noise recently and it seems like there are quite a few places in the image processing pipeline where noise can be introduced. Are manufacturing and design improvements that reduce read noise partly responsible for increasing DR? (Aside from features like dual gain output, od course.)
And re
ver/under - are you saying that highlight and shadow allocation on the curve is completely arbitrary? Isn't there some absolute relationship between the brightness of middle gray in the exposed image and where it falls on the manufacturer's curve, at least under the definition of the stated ISO?