Dynamic Range

[Graeme Nattress]

Dynamic range is a darn tricky thing to measure. You've got to have an image target of greater dynamic range than what device you want to measure. With the Stouffer transmissive stepwedge we use, it has 41 steps, each about 0.1 OD more dense than the previous (0.3 OD difference is 1 stop). That gives us, if you light it evenly and ensure no light gets to it from the front, a target of over 13 stops.

Next, everyone has their own method of measuring dynamic range. I've shown my methodology in this thread here: http://www.reduser.net/forum/showpost.php?p=89624&postcount=43

But some people like other tools, like Imatest, so above is the same chart through Imatest, showing a headline figure of 11.4 stops. The curve is a bit funky because I don't know how to get Imatest to interperet linear light data, which I find gives the most accurate results, and instead had to use a curve to get it to register the range correctly. Now, a curve cannot actually increase the dynamic range of an image, but it could, if done wrong, reduce the range. I suspect with more tweaking of the gamma curve I could get a better result.

Graeme

 

[number6]

Graham, just curious, will all cameras be tested to insure that all sensors are uniform? (And if so, what do you do with the over-achievers?)

 

[Graeme Nattress]

Sensors are tested and calibrated for uniformity.

Graeme

 

[Steven Parker]

I suspect with more tweaking of the gamma curve I could get a better result.

Graeme

Very interesting, and quite up-front of you to post the screen grabs.

But how do you mean tweak the gamma curve? Gamma adjustment in the test camera, or gamma adjustment in the testing software?

Forgive my ignorance; every day this Red thing pushes me further into new areas. Supersweet, but a bit overwhelming sometimes - now I am supposed to know not just light and color (and blocking and story arc and lenses etc etc) but also video engineering!

inch:

 

[Kevin Halverson]

Just a bit curious. What is the color temperature of the illumination source used in your test and where in the range of available analog gain was this test performed?

Thanks for the posting. Good confirmation of your methodology.

 

[Graeme Nattress]

Well, after decompression of the R3D file, we get totally linear light data. Putting this direct into Imatest didn't work right - give bizarre results when it tried to figure an inverse gamma curve, so I went with REC709. We don't by default adjust the shadows, but I had to do this to get the program to recognise the darkest data. Because we shoot RAW, there's no one way to "develop" that into an image. I don't know precisely what Imatest expects, so I had to fiddle with the curve controls in REDAlert! a bit.

Colour temperature was a ~5000k (WB picker reads 5446k) and analogue gain was camera standard.

 

[jbeale]

Just a bit curious. What is the color temperature of the illumination source used in your test and where in the range of available analog gain was this test performed?

I'm curious about that too. While a raw data file may not have a color temperature, is there some color temp. light for which the luminance S/N is maximum, hence DR also?

Maybe 5000K is some kind of optimum for the Red? I notice for example the R,G,B channel noise levels look very similar on Graeme's screenshot, whereas the blue-channel noise on the TIFF from the DCS test I analyzed was a lot higher (maybe due to 3200K light?) see below to compare.
http://www.bealecorner.com/video/RED/Step-Sept15/

At any rate this is an impressive demonstration. It does look to me like getting a good comparison involves more "hands on" work than I had realized, because the Imatest DR measurement program is apparently sensitive to the gamma curve in the shadows.

 

[Graeme Nattress]

Probably get best results in daylight, but native balance is around 5000k.

Graeme

 

[Stacey Spears]

The trick is to know the gamma curve that Imatest expects.

Graeme, have you tried exporting 32-bit vs. 16-bit images? I usually deal with 32-bit float when I process, and store, in linear light. With 8 - 10-bit original data, I am happy with 16-bit, but with a 12-bit A/D in the camera, I would be tempted to use 32-bit.

To make a long story, longer, would it be possible to get a 32-bit output option in RedCine?

 

[Graeme Nattress]

The sensor is 12bit linear, so, after demosaicing, 16bit linear has more than enough room to properly store, with enough precision, the image data. I don't, therefore, see any advantage in 32-float. However, I think REDCINE does EXR, which would be float?

Graeme

 

[Kevin Halverson]

Colour temperature was a ~5000k (WB picker reads 5446k) and analogue gain was camera standard.

Does this mean that the gain was at a mean or minimum? Is there perhaps an even lower analog gain value that could acheive even better S/N ratios?

Either way these are very impressive results!

 

[Darren Orange]

This is Redcode RAW...does this mean that RAW will be Greater than 11.4? As result of 10 bit vs 12 bit?

 

[GlennChan]

Uh... I see some nasty aliasing artifacts in that picture.

Does this mean that the gain was at a mean or minimum? Is there perhaps an even lower analog gain value that could acheive even better S/N ratios?

From my reading of the ISO threads (where Jim Jannard was posting the stills demonstrating very very high ISO), the Red team tested various levels of analog gain. Presumably the camera default yields the highest amount of dynamic range???

 

[Evan Owen]

Uh... I see some nasty aliasing artifacts in that picture.


From my reading of the ISO threads (where Jim Jannard was posting the stills demonstrating very very high ISO), the Red team tested various levels of analog gain. Presumably the camera default yields the highest amount of dynamic range???

There isn't any analog gain, AFAIK.

"Gain" (or ISO) is simply an adjustment of the RAW image captured directly from the chip. That's why it doesn't matter how the ISO is set when you shoot, only what the aperture is set at.

 

[Jannard]

Uh... I see some nasty aliasing artifacts in that picture.


From my reading of the ISO threads (where Jim Jannard was posting the stills demonstrating very very high ISO), the Red team tested various levels of analog gain. Presumably the camera default yields the highest amount of dynamic range???

We did not change any gain settings in the very, very high ISO shots. They were just stopped down, and down per meter readings and "push processed".

Jim

 

[jbeale]

I guess that would be the ASA 32000 example here: http://www.reduser.net/forum/showthread.php?p=58491#post58491

If we assume the Red is normally ASA 320, then 32000 is about 6.6 stops underexposed. At first I thought that was pretty unlikely without extra analog gain, but I tried a quick chart test with my 20D DSLR set to ASA 100 shooting raw at 6 stops underexposure and was able to pull out something visible, although with more chroma noise in the darker areas. Set to ASA 320 of course my camera would be even worse. I'd say that all these DR tests agree, Red is pretty impressive any way you look at it.

Edit: I just re-read you lit the 32000 shot for ISO 250, so that is 7 stops. I was wondering though, since I'd heard 320 to 500 recommended, can you actually use the camera at ISO 250 ? Maybe in some situations?

 

[Jannard]

Yes... ISO 250 is one of the choices in the menu.

Jim

 

[Emmanuel Cambier]

Yes... ISO 250 is one of the choices in the menu.

Jim

That's great news, is it already available in the shiped cameras ?


Emmanuel

 

[Graeme Nattress]

This is Redcode RAW...does this mean that RAW will be Greater than 11.4? As result of 10 bit vs 12 bit?

REDCODE RAW is 12bit though, and hence this is the same as you'd get for raw, raw RAW.

Graeme

 

[Brook Willard]

That's great news, is it already available in the shiped cameras ?


Emmanuel

Yeah, it's there.