Meaning of 16.5 stops of Dynamic Range

[Scott Crawley]

what excited me about the dragon sensor was that 2000 ISO will be as clean as 800 ISO on MX, that doesn't seem to be the case. I would happily trade resolution for a cleaner image.

You can. Shoot 6k and down sample to your delivery format of choice for a cleaner image.

 

[egon touste]

well the original R1 was specified as 11.3 I believe .. most of my clients rated it at 8 - 8.5 .. unless you want to measure using 2D graphs of charts..

so im guessing the dragon has 13 stops of usable..until I see different..

 

[Brian Boyer]

Bits to stop aren't directly related. That was precisely my point. You can have a 32 bit gradient or an 8 bit gradient and they will look subjectively similar if they go from black -> white. However when you are talking about image synthesis whether that's a simple photoshop gradient or else a 3D rendering, since there is no noise there is a direct relationship. 0.002 is one stop more than 0.001 therefore your floating point precision directly translates into how many steps, or stops there are. 2^10 bits = 2^10 steps = 10 stops. So the takeway isn't that 16 stops is 16 bits, it's that if you have 32 bits of image data in a 3D rendering, it'll look perfectly normal, even though it mathematically by definition has 32 stops of precision. You don't need a 32 stop display to appreciate an image with 32 stops just like you don't need a 32 bit display to appreciate an image with 32 bits.

Your mention of this was the first time I've ever heard anyone make this claim, so forgive my slowness to pick up what you're laying down. Maybe I'm just generally slow, but...

.002 is twice as much as .001. But, it seems to me that .002 is only one stop more than .001 if .002 is twice as bright as .001, and .004 is twice as bright as .002, etc.

Otherwise, it seems as arbitrary as saying $10 is one stop more than $5.

I get how the bit relationship works with the A/D converter, as it's working with brightness values derived from actual photons.

I don't get how a black to white gradient drawn in an 8-bit Photoshop document shows 2^8 stops. There's 2^8 steps of precision between the two values, which would make it more subject to banding than 2^10.

I don't think Photoshop internally views the brightest white value as 2^8 times as bright as the black one. Maybe that's exactly what it's doing. A 3D program that understands physically based lighting models might make that distinction but I'm not so sure Photoshop does.

I see how 10 stops of brightness can be expressed as 2^10 steps. What I don't see is how 2^10 steps can be said to equal 10 stops, unless there's doubling or halving relationship in the luminance values. Stops describe brightness in a specific way, otherwise it's just brighter (or more).

It seems to me all you can say is there's the same amount of numerical precision, when digitizing, in 2^10 stops and a Photoshop generated image with 2^10 steps. But saying 2^10 steps necessarily = 2^10 stops seems wrong to me.

"32 stops of precision" seems like a conflation of terms and concepts.


EDIT:

Maybe they are the same and I'm just not getting it. I've been screwing around in Photoshop and I can see how what you're saying seems to follow.

But, given the black to white gradient example, is the white in the 16-bit Photoshop document brighter than the white in an 8-bit document or is it just that it has a higher RGB value because there's more values to describe the range from black to white in the 16-bit version?

If what you're saying is true, it seems like it should be brighter, unless Photoshop automatically always assumes the brightest white theoretically possible (or, say, up to 64 bits) when making calculations, even in an 8-bit document. I think I'm confusing myself at this point.

I just don't want to make the mistake of coming to the wrong conclusion about what the numbers represent based on the similarity of how they're expressed.

 

[Scott Crawley]

Gunkeik warned that this was going to get academic, and it has.

The most important aspect of improved DR for me is that I can capture a scene of higher contrast and retain more detail at both ends of the exposure. It means I don't have to work as hard to reduce the contrast range inherent in the scene I am trying to capture. 3 stops more or 1, it is more DR and more DR is good. :drool:

 

[Dunkan Wolf]

It's just kind of sad that Red needs to fumble around with numbers rather being upfront about their specs. Are they really that insecure about their market position?

I'm not saying this to bash Red. i'm saying this because Red's reputation isn't too good with alot of people i meet, which can be frustrating.

 

[Gunleik Groven]

It's just kind of sad that Red needs to fumble around with numbers rather being upfront about their specs. Are they really that insecure about their market position?

I'm not saying this to bash Red. i'm saying this because Red's reputation isn't too good with alot of people i meet, which can be frustrating.

Hm... This was a fun and contradictive post...

re-read.

 

[Jan Reiff]

MX - Dragon, same lense, red prime, complete open @ 1.8 in high noon sun, no ND


R3D still here

http://we.tl/aIMG7FKcpE

let me know what you think ...

 

[Jan Reiff]

It's just kind of sad that Red needs to fumble around with numbers rather being upfront about their specs. Are they really that insecure about their market position?

I'm not saying this to bash Red. i'm saying this because Red's reputation isn't too good with alot of people i meet, which can be frustrating.

you should meet more people like Fincher / Scott ;-)

 

[Matthew Scott]

It's just kind of sad that Red needs to fumble around with numbers rather being upfront about their specs. Are they really that insecure about their market position?

I'm not saying this to bash Red. i'm saying this because Red's reputation isn't too good with alot of people i meet, which can be frustrating.

EXACTLY I'm not actually concerned about how good or bad the dragon is, I'm concerned about RED advertising facts about new release products that don't seem to add up. Then to top it off, many already-dragon-owners seem to think that that's just fine, there's nothing to worry about, the camera is awesome.

 

[Mark Wilkinson]

That's a brutal torture test. I don't suppose you grabbed any frames at other exposures? It might teach us a bit more - after all, you can clip any sensor if you try hard enough.

 

[Stephen Matthews]

Useable DR is a very subjective thing each person has their own cutoff, and in my opinion there is no universal DR test that is immune to human error.

There is a huge difference between comfortable working range to actual usable DR and technical specifications DR, yes RED's 16.5 at this point may just be a lab number and no one would push the image to those extremes but it's marketing that's how it works, can you achieve that in the lab? yes, then put it on the box!

Everyone does it just when red does it's time to stop the world and let them know. Now I'm not going to lie to myself about what it can do but to be honest at these DR amounts with proper understanding of light it would be embarrassing if you couldn't make this sensor shine. If I was prepping for a major shoot I wouldn't care less about marketing numbers on a web store, I would know my real world numbers for the camera and what it can do and if I didn't I wouldn't be doing my job right.

Seems like the same old same old to me, manufacture makes out there claim I have a hard time believing so I get my hands on what I can and push it as much as I can and if that doesn't appease me I get my hands on a camera and run it through the paces myself, all part and parcel to the work we do.

I guess it's just a bunch of nerves, as people prep to get their camera's, defend the purchase, or defend the non purchase, either way kind of a non issue for me, numbers are meaningless, real world functions and realities are all that matter.

It's called propaganda people they just changed it to marketing.... watch century of the self if you would like some further viewing on the subject, interesting tale on post war consumerism and the use of Freud's work to create a culture of consumerism.

Have good one! and keep watching..... the stars?

 

[Mark Wuerthner]

Would be nice if Phil or John M could post a shot (real world) that shows 16.5 stops of DR. It would need to be an R3D that we can play with and not a processed image that they colored and noise reduced. I think that would settle the debate. If we can see 16.5 stops then great, if not, well then we know for sure you cannot capture 16.5 stops in a standard shot from the camera. 14.8 stops according to DXOmark is great, I think we just need a solid number that either refutes that claim by bettering it or that is our solid number and end of discussion. I don't consider digging out noisy blacks a part of the total DR since it requires major post work to be usable. Same for extreme highlights that need to be dialed back. My thought is if the claim is 16.5 stops, then it should be available in a single shot without any corrections in post. That means above the noise floor and below the clipping point.
There is a simple way to shoot a very wide dynamic range of brightness values, a way that is very precise and is also very easy to understand the results. DSC Labs makes a backlit chart named Xyla that has 20 stops of dynamic range in 1 stop increments. Aim the camera at the chart (in a completely dark room), roll camera, slowly adjust aperture from wide open to closed down, put footage into post application, freeze footage when brightest chip clips on waveform monitor, count grayscale chips down from that. Here is a perfect example of this process:

http://provideocoalition.com/aadams/story/alexa_dynamic_range_its_all_in_how_you_use_it/P2

Here's another example of it:

http://blog.abelcine.com/2011/03/31/abelcine-tests-the-sony-f3-dynamic-range/

 

[Jan Reiff]

another test between MX and Dragon

https://flic.kr/p/ndTTyD

first let me say, i think some expect TOO much - MX sensor was and is a very good sensor (!!) capturing lots of information in RAW - i think we shouldn´t talk here as MX is an old and unusable camera ... it´s a camera Blockbusters are shot on.
Dragon is better in the whole image impression - and in the details you have to discover for yourself.
here is a comparison of a shot of this beautiful 1994 Porsche in brightest sun of the day. Unfortunately, i can´t load the Raws, as someone of the team format the SSD of the MX today :-(
as you can see - the Dragon separates bit better (clearer) in the shadow (i admit this is ADD) and the roll off is better ... and - the Dragon is f16 and MX is closed at f22 (Red Prime - at f22 its going to be bit soft at MX, but of course Dragon IS sharper)
It´s up to you what you see in all this, 1 stop more, 2 stop more, ... my personal conclusion after 2 days testing MX / Dragon is:
Dragon is clearer in every aspect, it´s like you take off a filter from the MX. It´s brighter, more shiny. MX f11 for example is f8 at the Dragon. In the shadows colors and separation of the tonal values is clearer, less chroma noise.
Maybe someone expected more in DR ... don´t know, but wait and see the whole Image, see the improved roll off (for me, at least, like half a stop more in this aspect) and then judge ... and, not to forget - we have a 6K image here, not 2, not 3, it´s 6K, ...


Cheers!

 

[Gunleik Groven]

EXACTLY I'm not actually concerned about how good or bad the dragon is, I'm concerned about RED advertising facts about new release products that don't seem to add up. Then to top it off, many already-dragon-owners seem to think that that's just fine, there's nothing to worry about, the camera is awesome.

Curious...

What exactly is wrong about liking a camera you like?

As stated earlier, I actually didn't even plan to do the upgrade, but changed my mind after using it.

What is worrying about that?

RED definitely did not threaten me to be enthused...

 

[Dunkan Wolf]

Curious...

What exactly is wrong about liking a camera you like?

As stated earlier, I actually didn't even plan to do the upgrade, but changed my mind after using it.

What is worrying about that?

RED definitely did not threaten me to be enthused...

Nothing. But re-read his post.

 

[Corey Robson]

I'm going to regret posting this but here you go...

CF Dragon - new OLPF - build 5.2.14 - Lens - 75mm Cooke S4i, 5600K, Dragoncolor/Redlogfilm, captured from brain HD-SDI to Atomos Samurai Blade using ProresHQ.

Alexa - Shot this last year so it was running SUP6. Shouldn't affect DR much if at all. Lens - Cooke 50mm S5. 5600K, Log-C recorded to SxS using ProresHQ.

Chart - Gamma&Density Trans-1 16stop DR test chart - backlit by 5600K Rosco Litepad Axiom (I taped off the brightest pluges partially and flagged off half of the chart to help reduce flare contamination)

Waveform captures made from my Leader 5330.

Methodology - I set the ISO on both cameras as indicated on the stills below and shoot from wide open to T22. Later I scrub through the clips on my Mac Pro using an AJA Kona 3G outputting to the 5330 waveform. When the brightest pluge just begins to fall from it's peaked level I take a frame grab. No adjustments have been made to any of these clips. I did not have access to an ARRIRAW recorder when I tested the Alexa. Obviously R3Ds and ARRIRAW clips can be manipulated further than PRHQ. This was just to test what I get straight from the Dragon brain using Redlogfilm at 1080 or Log-C out at 1080 on an Alexa.

This is not the definitive DR test some of you are hoping for but you may find it useful in some way. I shoot most cameras I work with using this chart as it gives me a common place to start from before making exposure decisions. Take from it what you will, I'm personally not interested in debating which has greater DR as they are both awesome. Plenty of room there to create your art.

Below - Dragon - 400ISO

Below - Dragon - 800ISO

Below - Dragon - 1600ISO

Below - Alexa - 400ISO

Below - Alexa - 800ISO

Below - Alexa - 1600ISO

Links to full res 1080 TIFFs of each:
https://dl.dropboxusercontent.com/u...00ISO_DCRLF_A006_C009_0506ND.mov.Still001.tif
https://dl.dropboxusercontent.com/u...00ISO_DCRLF_A006_C016_0506SF.MOV.Still001.tif
https://dl.dropboxusercontent.com/u...00ISO_DCRLF_A006_C012_0506E4.MOV.Still001.tif
https://dl.dropboxusercontent.com/u/55061736/DRAGON_DR_TEST/ALEXA_400_5600_1.mov.Still001.tif
https://dl.dropboxusercontent.com/u/55061736/DRAGON_DR_TEST/ALEXA_800_5600_1.mov.Still001.tif
https://dl.dropboxusercontent.com/u/55061736/DRAGON_DR_TEST/ALEXA_1600_5600_1.mov.Still001.tif

NOTE: I have to investigate the cause of the faint magenta flare half way up the frame on the Dragon pics linked to above. Might be something in front of the sensor that needs addressing...

 

[Clayton Burkhart]

The above charts seem to be intdication 14 stops for the Alexa, 13 for the Dragon.

 

[Nick Morrison]

Obviously R3Ds and ARRIRAW clips can be manipulated further than PRHQ. This was just to test what I get straight from the Dragon brain using Redlogfilm at 1080 or Log-C out at 1080 on an Alexa.

What effect on measurable DR does exporting to 10bit Prores have vs looking at the 16bit RAW? From what I understand, Alexa was designed innately around a prores workflow (since augmented with RAW), whereas RED is designed around RAW workflow. Curious how test would have looked with both shooting "natively", or even RAW.

However, if one was curious to see what a Red Proxy would look like vs Alexa directly, this is certainly a very informative test.

Thanks for sharing!

 

[Jan Reiff]

mmh, i´m not good at reading measure stuff and waves and so on, ... but in the stills of the dropbox, ... at 800 i see nicer highlights at the dragon to be honest, (especially at ISO400) there seems to be even "more" than Alexa has (Alexa seems closer to clipping) - in the end of shadows is a bit more separation at Alexa, but still at Dragon too, just not so clear
at ISO 1600, Alexa shows more in shadows than the Dragon Sensor, but the highlights at Dragon look nicer, ...
i heard and saw so many different results, ... i think it´s time to start shooting movies again, just in case one of the cameras has 1 stop more, who cares ? the real thing is the set and the shoot, not marketing of both companies make the images or film, we do ... and with both cameras you will achieve great images, ... with Dragon even in a 6K frame ;-)

 

[Hrvoje Simic]

In a single-shot DR testing route it would be helpful:

- Pointing cameras at a high contrast real life scene, containing greater range than camera(s) can capture
- Using spot meter for measuring values of distinguishable scene elements, from overexposed to underexposed ones
- Providing scene measurements
- Providing captured results

Additionally, providing results in RAW enables checking the usability of captured DR.