Exposing + Native ISO

[Mark L. Pederson]

I can't say for sure at this moment because I have not done my own testing in a wide variety of circumstances, but based on what I am seeing in R3D files, I'm thinking it might be a true 800 ISO sensor now.

I'm not talking about where RED puts the ISO as a default, but where it really sits. Maybe also rather than go with ISO, EI would be better.

Battistella

Okay - but again - with respect to "where it really sits" - if we all agree that = same DR above 18% grey and same DR below 18% grey - HOW do you define the DR of the lower stops? Because I tolerate a lot less noise than other folks - so - what would the technical process be for defining the lower stops of DR?

When you do your tests, how exactly will you decide how many stops are below 18% grey?

That is what I am struggling with - maybe I'm missing something simple - if so - please advise

 

[David Battistella]

Like you, I'm going to favor less grain. When the blacks get too grainy, I won't want to go there, so for me it will come down to acceptable amounts of noise/widest DR. I'll probably end up near your numbers. The dragon stuff is very low noise, from what I have seen. This could be considered conservative and people will push back (as many who insist on 800 for EPIC, where many are in the 320-500 range). I've always been in the 320-500 as noise is the main factor for me.

battistella

Okay - but again - with respect to "where it really sits" - if we all agree that = same DR above 18% grey and same DR below 18% grey - HOW do you define the DR of the lower stops? Because I tolerate a lot less noise than other folks - so - what would the technical process be for defining the lower stops of DR?

When you do your tests, how exactly will you decide how many stops are below 18% grey?

That is what I am struggling with - maybe I'm missing something simple - if so - please advise

 

[Scott Crawley]

We are getting into terminology issues here, and in a way, I think that is not a bad idea.

Funny, that is where I was at the start, even in the false color thread. What's in a name? Everything. It turns out words have meaning! A definite meaning... Sometimes two. ;-)

Sorry, that mini rant was about American culture, not your command of the language.

 

[Mark L. Pederson]

Like you, I'm going to favor less grain. When the blacks get too grainy, I won't want to go there, so for me it will come down to acceptable amounts of noise/widest DR. I'll probably end up near your numbers. The dragon stuff is very low noise, from what I have seen. This could be considered conservative and people will push back (as many who insist on 800 for EPIC, where many are in the 320-500 range). I've always been in the 320-500 as noise is the main factor for me.

battistella

And THIS is the elephant in the room. You are asking for a technical specification that requires a subjective interpretation.

And this is why I keep teaching that it's a RANGE not an exact value.

I'm not being a jerk here - I would love for everyone to agree and feel warm and fuzzy - and I'm sure this thread is very helpful for some folks - but let's peel back another layer of the onion -

#1: scaling the image down - from 5K or 6K to HD or 2K - reduces the perceived noise. I have MANY times looked at a HD image from a MX and saw no perceived noise - but then when viewing in 4K/UHD - I could see more noise than I found acceptable.

#2: some cameras apply NR in the camera. RED does not. You CAN, obviously apply it in post. 2000 ISO low light images from Dragon stand up great with just a small amount of NR applied in post.

I have tested and proven - that taking a 6K image and scaling down to 4K/UHD AND applying a small amount of NR in post processing - holds more detail than scaling ARRI RAW 2.8K up to 4K/UHD. So ... it's not like NR is "cheating" - it's a real tool.

So ... obviously you don't want a "base ISO" for finishing in HD vs. 4K, right? Or a "base ISO" that strictly excludes NR post processing, right?

BUT ... in all seriousness - you can extend your dynamic range of your lower stops with NR post processing and scaling - and again - unless someone can show me some technical standard for how much noise is too much - it all ends up in a subjective interpretation - and that can be 1 full stop easy.

Again - I'm not trying to be a jerk here - if someone can explain to me how they would technically define how many stops of DR are below 18% grey I'm all ears.

 

[David Battistella]

And THIS is the elephant in the room. You are asking for a technical specification that requires a subjective interpretation.

YUP. I want RED to determine that. We can then be subjective to their interpretation with our creative choices.

And this is why I keep teaching that it's a RANGE not an exact value.

I would teach this too, because it is a creative tool. But there is a reference point set by the manufacture, Like a 500ASA sticker on a film can.

I'm not being a jerk here - I would love for everyone to agree and feel warm and fuzzy - and I'm sure this thread is very helpful for some folks - but let's peel back another layer of the onion -

#1: scaling the image down - from 5K or 6K to HD or 2K - reduces the perceived noise. I have MANY times looked at a HD image from a MX and saw no perceived noise - but then when viewing in 4K/UHD - I could see more noise than I found acceptable.

You are not being a jerk. This needs to be determined based on FUTURE deliver formats 4K and beyond, therefore I would say it more important to be conservative for eventual 8K blowups. Let's pretend HD is dead.

#2: some cameras apply NR in the camera. RED does not. You CAN, obviously apply it in post. 2000 ISO low light images from Dragon stand up great with just a small amount of NR applied in post.

Again here, RED just needs to deliver the RAW image and people should sharpen to taste. Always believed this is the 100% correct direction by RED, an honest image.

I have tested and proven - that taking a 6K image and scaling down to 4K/UHD AND applying a small amount of NR in post processing - holds more detail than scaling ARRI RAW 2.8K up to 4K/UHD. So ... it's not like NR is "cheating" - it's a real tool.

So ... obviously you don't want a "base ISO" for finishing in HD vs. 4K, right? Or a "base ISO" that strictly excludes NR post processing, right?

Has to be excluded. Would Kodak consider how one person sharpens in LA compared to how they sharpen in NY? or Hong King? No. RED delivers the picture based on the future, not now. I son't agree with picking an base exposure based on HD finishes, but i would agree with one based on 4K or 8K finishes.

BUT ... in all seriousness - you can extend your dynamic range of your lower stops with NR post processing and scaling - and again - unless someone can show me some technical standard for how much noise is too much - it all ends up in a subjective interpretation - and that can be 1 full stop easy.

Again - I'm not trying to be a jerk here - if someone can explain to me how they would technically define how many stops of DR are below 18% grey I'm all ears.

Post tools should not be RED's concern, unless they were building that functionality into their own exposure tools. I never would sharpen with RED's tools in RCX as they produce little black artifacts in prints, etc. (that really needs to get straighten out)

Why? Because we can't predict that is the way it will be. They have to focus on the IMAGE CONTAINER, what goes into the R3D, that's it. Like a film shot 50 years ago scanned today at 6K is going have the benefit of tools that were not available at the time. But the container (the film) reveals more than it did or could then, compared to now.

RED should pack all they can into that container so that in 50 years the R3D's we extract will blow our minds based on how we can decode it then.

It's all about down the road, HD is in the rear view mirror.


Battistella

 

[Mark L. Pederson]

Post tools should not be RED's concern, unless they were building that functionality into their own exposure tools. I never would sharpen with RED's tools in RCX as they produce little black artifacts in prints, etc. (that really needs to get straighten out)

Battistella

Well ... I am pretty sure that Kodak and Fuji took post processing into their ISO determination for film stocks.

You say "I want RED to determine that" - can you tell me, technically and subjectively, how you want them to "determine that"?

So ... RED picks a number - Phil picks a different number - and I pick a different number - they are all close - but all different.

Sounds like a RANGE to me.

But okay - since it's the manufacturer declaring it - it's "right" then?

I'm NOT saying RED should not just pick a number in the RANGE. But if you are gonna use that as your BASE ISO - then please enlighten me, and RED, how they should determine that number so that guys like me won't say - that's not the right number - the REAL number is XYZ Because THAT is exactly what is going to happen.

IMO - threads like this - and teaching people that "Native ISO" is really a RANGE not an exact number - is better for everyone.

But Adrian's comments and arguments are awesome and valid.

That said - I think we can squeeze that RANGE down a bit - make it more narrow. Right now - I'm gonna say Dragon it's between 800 and 950 - and again, in MOST situations - my preference will be to rate it at 640 - but I will wait until I get a new OLPF and then I will re-test EVERYTHING and I will tell you what the smallest RANGE I feel comfortable defining as "NATIVE ISO" - specifically for 4K finishing.


Good thread, huh

I love Reduser. I'm glad to be back.

 

[Mark L. Pederson]

After more reading - The ISO 12232:2006 standard


"The ISO standard ISO 12232:2006[55] gives digital still camera manufacturers a choice of five different techniques for determining the exposure index rating at each sensitivity setting provided by a particular camera model. Three of the techniques in ISO 12232:2006 are carried over from the 1998 version of the standard, while two new techniques allowing for measurement of JPEG output files are introduced from CIPA DC-004.[56] Depending on the technique selected, the exposure index rating can depend on the sensor sensitivity, the sensor noise, and the appearance of the resulting image. The standard specifies the measurement of light sensitivity of the entire digital camera system and not of individual components such as digital sensors, although Kodak has reported[57] using a variation to characterize the sensitivity of two of their sensors in 2001."

and

"Noise-based speed[edit]
Main article: Signal to noise ratio (imaging)


Digital noise at 3200 ISO vs. 100 ISO
The noise-based speed is defined as the exposure that will lead to a given signal-to-noise ratio on individual pixels. Two ratios are used, the 40:1 ("excellent image quality") and the 10:1 ("acceptable image quality") ratio. These ratios have been subjectively determined based on a resolution of 70 pixels per cm (180 DPI) when viewed at 25 cm (10 inch) distance. The signal-to-noise ratio is defined as the standard deviation of a weighted average of the luminance and color of individual pixels. The noise-based speed is mostly determined by the properties of the sensor and somewhat affected by the noise in the electronic gain and AD converter.[55]"

"Despite these detailed standard definitions, cameras typically do not clearly indicate whether the user "ISO" setting refers to the noise-based speed, saturation-based speed, or the specified output sensitivity, or even some made-up number for marketing purposes. Because the 1998 version of ISO 12232 did not permit measurement of camera output that had lossy compression, it was not possible to correctly apply any of those measurements to cameras that did not produce sRGB files in an uncompressed format such as TIFF. Following the publication of CIPA DC-004 in 2006, Japanese manufacturers of digital still cameras are required to specify whether a sensitivity rating is REI or SOS."

180 DPI, huh?

Cough ... Cough ....

 

[David Battistella]

But okay - since it's the manufacturer declaring it - it's "right" then?

It's not right or wrong. It's the starting point. Take vision stock (amazing stock) Kodak came up with an EI for that film. We can argue all we want, but they determined that, that did not make them right, but it did make them the company that said, "we made this and we think this." now people can disagree, but it's not up to me to determine that for Kodak, or for RED.

I'm NOT saying RED should not just pick a number in the RANGE. But if you are gonna use that as your BASE ISO - then please enlighten me, and RED, how they should determine that number so that guys like me won't say - that's not the right number - the REAL number is XYZ Because THAT is exactly what is going to happen.

Of course it will. But do you hear a bunch of talk around the ALEXA base rating? Not really because when people took out their meters and starting lighting the way they did with film, they were ending up with consistent results. With RED it's been 800-320-500, but that talk was there because metered results vary. Now, going back to Evin's test, you can see dragon is responding based on the ISO you set your camera and Meter too. That is great news, so I think these are old issues based on how things have "needed" to be done now.

IMO - threads like this - and teaching people that "Native ISO" is really a RANGE not an exact number - is better for everyone.

But Adrian's comments and arguments are awesome and valid.

That said - I think we can squeeze that RANGE down a bit - make it more narrow. Right now - I'm gonna say Dragon it's between 800 and 950 - but I will wait until I get a new OLPF and then I will re-test EVERYTHING and I will tell you what the smallest RANGE I feel comfortable defining as "NATIVE ISO" - specifically for 4K finishing.

Good thread, huh

I love Reduser. I'm glad to be back.

I agree. It's a great thread. Constructive and getting back to the way it was at the start. If I had a dragon in my hands I would probably be giving you better arguments, but I don't even like to argue, I think anything that from RED that removes ambiguity is great. Then we are free to be ambiguous about all this stuff and that stimulates discussion and creates an exchange of ideas. That gets rid of fanboyism and creates community.

Battistella

 

[Patrick Tresch]

Thanks Adrian,

Best visual explanation I've seen so far. Cool would be to have a real stripe from the actual Dragon sensor to get the noise feeling. Like in the old days a densitometry stripe. Or add a red noise stripe in the blacks between -5 and -9. So people understand that the blacks at iso 3200 aren't as clean as at iso 160. At iso 800 with the "old olpf Dragon" I see more like +6 -8 stops.

Pat

 

[David Battistella]

180 DPI, huh?

Cough ... Cough ....

This is good reading.

A full frame tiff from a 6K stream out of RCX is...wait for it....72 DPI. So, they rate based on 2.5 times the DPI of computer monitor resolution.


Battistella

 

[Terry VerHaar]

As we pause to take a breath here, I hope you gentlemen don't mind fielding a question from a very studious lurker on this thread.

I learned the use of film over about three decades as an amateur still photographer (including many courses and books), which informs my RED shooting, although I have never been a film cinematographer. So, the concepts of ISO and proper exposure are well seated within me. One thing I have had to deal with in shooting RED is the somewhat counter-intuitive (for me) sense of changing ISO to shift the middle grey. I love knowing that changing ISO doesn't change the light gathered by the sensor (and that RED does no signal gain). To be honest, I have pretty much ignored ISO, shooting at 800 nearly always. I occasionally use high ISO just to make the image more visible in my monitor but I am careful not to let it influence my exposure, instead relying on goal posts and traffic lights.

Now, it does seem like this discussion and other information from RED, that shifting ISO can help me get better exposure but again, for me, in a counterintuitive way; meaning that a more high key scene would call for setting a higher ISO, to protect highlights, while a darker scene might call for lowering ISO to guard against excess noise (although this half of the equation seems harder for me to grasp).

How far off am I here? 180 degrees? LOL (I would love to grasp one more tool for getting proper exposure on my Scarlet - soon to be Epic.)

Thanks for indulging me... :smiley:

 

[David Battistella]

Ansel adams ZONE System is your friend Terry because it relies on, amonst other things where you place middle grey in a given scene. So you approach exposure on a scene basis and where the middle values are in that scene.

Applied to shooting situations you would move your ISO rating up or down to set that middle grey point.


With a film ISO basis you have a good grasp.

Battistella

 

[Mark L. Pederson]

Now, it does seem like this discussion and other information from RED, that shifting ISO can help me get better exposure but again, for me, in a counterintuitive way; meaning that a more high key scene would call for setting a higher ISO, to protect highlights, while a darker scene might call for lowering ISO to guard against excess noise (although this half of the equation seems harder for me to grasp).

How far off am I here? 180 degrees? LOL (I would love to grasp one more tool for getting proper exposure on my Scarlet - soon to be Epic.)

Thanks for indulging me... :smiley:

You are singing my song Terry. You are not off at all. Again, it comes down to your exposure METHOD - and most methods factor ISO into the calculation.

 

[Terry VerHaar]

Yes, David, I was always impressed with sophisticated simplicity of the Zone System. It made a big impression on me.

Mark - thanks for opening my eyes to the use of the Exposure Tool with this thread. I wasn't aware that it had ever been based on the raw image so I tended to ignore it. When they get it fixed, it will be a great tool. And, thanks for confirming my suspicions about the use of ISO - it's always hard to go with your "counter intuition" without such confirmation.

Great thread!

 

[Zeb B]

Attn RED Online Education: over 250 thoughtful (yet sometimes conflicting) posts on this topic in 3 days. Clearly a subject for a series of 10 minute videos complete with infographics, charts and light meter use to explain this obviously confusing but very critical topic. RAW/ISO/DR are the keys to the RED kingdom but it seems like maybe 20% of RED Owners have a firm grasp on how to exploit and manipulate these key RED tools.

Take the top 5 most knowledgeable and experienced guys in this thread and compile their points filtered through what RED knows to be true. Hire the awesome presenter Mark Wallace to script edit / produce and film the info. I'll fly to AZ and loan him my Dragon for the shoot. BAMM done!
http://www.snapfactory.com/live/
http://blog.snapfactory.com/

Education is the key to RED's longterm growth - keeping knowledge locked up in thousands of random RU pages brings viewers to a website and geeks chatting but does not effectively promote RED's key advantages to those who are ignorant.

Do you want pages views or camera sales through industry understanding and acceptance?
Educating those 99% in the industry who misunderstand (and therefore talk shit about) the RED advantage is the key - not those loyal fans who camp out on RU eating tech chat for breakfast.

 

[Mark L. Pederson]

Yes, David, I was always impressed with sophisticated simplicity of the Zone System. It made a big impression on me.

Mark - thanks for opening my eyes to the use of the Exposure Tool with this thread. I wasn't aware that it had ever been based on the raw image so I tended to ignore it. When they get it fixed, it will be a great tool. And, thanks for confirming my suspicions about the use of ISO - it's always hard to go with your "counter intuition" without such confirmation.

Great thread!

I think the graph that Adrian posted - (ironically pulled from the Alexa website) - really helps people understand how the dynamic range shifts relative to 18% grey. And yeah ... it's a bit counter to think that you might want to increase your ISO when you go outside

 

[Hrvoje Simic]

Except all the other cameras let you change ISO in the camera and it's how lot's of folks expect to work.

ISO value has been inherited from film realm to give a point of reference to cinematographers used to working with film. Using the similar methodology helps their accustomed workflow.

Experienced cinematographer is aware of effects and repercussions of "pushing" ISO. Shooters not familiar with same concepts may easily see it as just another setting unaware of the repercussions.

Red has chosen to cater different market segments where expectations and user experience greatly differ. Knowledgeable shooter knows exactly what ISO push does. Less experienced people can perceive ISO as "this setting gives me more light with less lighting and with WYSIWYG it looks okay."

Mental process: "I increased the sensitivity, no biggie, NR is going to cover the cost"
Reality: sensitivity is the same, you just re-distributed the density to upper parts of the histogram, reducing it in lower parts and NR is covering just one part of the cost

And it may look okay right there until the material gets to a post room with controlled light conditions and image is shaped closer to desired outcome by manipulating contrast hue and saturation values, when shadows start to reveal goblins in a rave party and little Mary turns to a plastic puppet.

Underexposing a skintone for 2 stops may seem as "not a big deal" on a slider scale but many are not aware of (or are just taking too lightly) the fact that 2 stops difference is 4 times amount of light. In cases where particular camera issues like usable DR are debated there seems to be a lack of comprehension of exponential scaling which a human brain compensates by default to allow the perception of vast range of information which surrounds it. I have seen many examples proving that and attempts of digging out the detail from highly underexposed footage. Talking just about the relatively solvable noise issue with color quality and lack of data being disregarded.

No matter how much cleaner shadows are, and regardless of RAW or bits used or available DR, if particular part of the image is greatly underexposed the usable data is not there, the well is not filled enough. Skintones turn to butterfly's wings, shadow work turns to shoveling gravel and post work is a mess. I am sure you and your clients are well aware of that, but many times this gets overlooked.

I would also say that many people seem to be chasing the desired "look" based on inadequate techniques in post and/or trying to compensate for inadequately shot material, both routes being based on misconceptions and unrealistic expectations and relying on camera's abilities to compensate own mistakes.

One of misconceptions which seems to exist is the misinterpretation of ISO. Based on mistakes like these many times the imaging apparatus gets blamed after people are perceiving cuts and bruises they themselves caused, unaware of how much they have mistreated the material or image acquisition process.

Shooter pushes the ISO, than additional material-pushing "little" tweaks, than a strong curve on top of all that, then we have threads about poor color rendition and splotches and whatnot. Plus endless Alexa vs. Epic threads and experienced users fighting windmills trying to explain all the factors over and over again, with valuable info being buried in forum threads in constantly repeating topics.

In this example, for the illustrative purposes let's say base is 800, - labeling: +1push (ISO 1600), +2 push (ISO 3200) might be more informative for the shooter without allowing the assumption of "variable ISO without repercussions". Just a little naming convention tweak for camera and post. Combined with 3D LUT support cinematographers are less likely to have unpleasant surprises in post. Just throwing ideas...

as long as someone can explain to me how we judge/define what is dynamic range at the bottom with respect to noise, etc. - that is what I'm still unsure of - please advise -

I can offer one potential route suggestion.

This currently very relative point could be established by joint testing and definition by Red and a group of experienced cinematographers, finding a middle ground between technical and aesthetic, between exact and subjective, between quantifiable and felt.

It would be beneficial that this group consists of cinematographers with differing criteria and tolerance, and that mutually agreed upon and officially stated reference point/range is subtly conservative one, thus avoiding most optimistic expectations and corresponding disappointments and endless debates which tend to follow them.

 

[Mark L. Pederson]

This currently very relative point could be established by joint testing and definition by Red and a group of experienced cinematographers, finding a middle ground between technical and aesthetical, between exact and subjective, between quantifiable and felt.

Or ... you can just accept that because of subjectivity to the noise floor - it's a RANGE

 

[Scott Crawley]

...

180 DPI, huh?

Cough ... Cough ....

Thanks for posting that ISO info. As for 180, it is mid-way between typical printing resolition 260 dpi and screen resolution 72 dpi. Half way is a fair place to meet.

 

[Evin Grant]

David, just for clarity's sake I was using the TMax-100 story to illustrate that Red's use of ISO was more accurate than Kodaks. I have used ISO 800 on the MX as my base exposure for the entire time I had that sensor. The only time I intentionally reduced ISO was when shooting VFX plates and green screen that I knew would benefit from the cleaner shadows. Specifically though the rated 800 ISO on the MX has been totally accurate to my meters since day one. I never tested to see if this was the mid point of the DR, but then that never seemed to matter to those that insisted that it was a 320-500 ISO sensor even though that was definitely not the mid point of the DR, which would have been closer to EI 1600 for 6.5 stops even distribution.

Oh and I used about 10 rolls for testing, the 100 rolls was for the entire class