Having trouble measuring the DR.

[Adam W Watson]

SOLVED: see my next post below to see where my mistake was.

Hello, I got a couple upgraded dragons in the other day and I've just started to test them to confirm their specs among other things, and I'm having trouble measuring the dynamic range. The problem is that I'm only coming up with 14 stops. The advertised spec is 16.5 yes? I've tried it with the log gamma and RAW view mode only to come up with the same results.

My method is to use a ChromaDuMonde chart with 6 stop grey gradient. I have the Red plugged in to a monitor with wave form and I have the "exposure" display activated on the red (the red, purple and green metering overlay mode). I light the chart so that the whitest chip on the chart is just blowing out, indicating red on the Red LCD. Then I look at the darkest chip and close the aperture down until it turns green, indicating middle grey. I didn't have to, it was already there, indicating 6 stops from middle grey to blown out. (an arbitrary aside, I noted that the green indication of middle grey was measuring at about 42 IRE) Then I turned down my light source until the white chip was showing as green (42 IRE). Then I close down the aperture until the black chip turns purple (no exposure). I had to close it down 2 stops for this to happen. Then I add it all up: 6 stops from middle grey to blown out, plus 8 stops from middle grey to zero exposure, equals 14 stops.

What am I missing? Has anyone else tried to measure this yet?

Also does anyone know what ISO level has been recommended by Red as the base ISO? I can't seem to find that information anywhere. I tried this test at 800 and 2000 ISO, both adding up to the same number of stops.

 

[John Marchant]

Purple is not zero- it's the beginning of the last 10% - and that last bit on dragon holds a lot of magic you're writing off.

i've measured the DR practically, against a meter, with a xyla -21 chart, and with my sekonic calibration chart. All come out at over 16. ISO makes no odds in the final outcome.

 

[John Marchant]

BTW, this is the easiest 16 stop test:- try exposing the centre of your chart just at clip at 1/50th shutter, f2.8

16 stops below that is f22, 1/51200 shutter (yes it goes that high- go to 'advanced' and enter it manually)

view both clips RLF in RCX and you should just see detail in both shots around the same chip on the chart

 

[Phil Holland]

Can you link me to the specific chart you are using? The Xyla 21 or DX-1 102-Db are a good way to measure this specific imaging system.

I would need to "be there" to fully see what you're doing to figure out if there's an error in the procedure.

If you are fully testing check out the values of REDgamma3, REDlogFilm, and Linear.

You will want to test the intermediate ISO values from 250-2000 individually to see where tonal balance and separation occurs. Base sensor sensitivity is one thing, "base ISO" is another. The total dynamic range should be visible on the chart within that full ISO range, though a variation of around .5 stops could happen if I'm recalling correcting depending on gamma space. Linear reveals the most useful information on that front at the moment. In practice and for practical shooting REDlogFilm is producing the best results for me.


** edit. John beat me too it!

 

[dio Zafi]

Hello, I got a couple upgraded dragons in the other day and I've just started to test them to confirm their specs among other things, and I'm having trouble measuring the dynamic range. The problem is that I'm only coming up with 14 stops. The advertised spec is 16.5 yes? I've tried it with the log gamma and RAW view mode only to come up with the same results.

My method is to use a ChromaDuMonde chart with 6 stop grey gradient. I have the Red plugged in to a monitor with wave form and I have the "exposure" display activated on the red (the red, purple and green metering overlay mode). I light the chart so that the whitest chip on the chart is just blowing out, indicating red on the Red LCD. Then I look at the darkest chip and close the aperture down until it turns green, indicating middle grey. I didn't have to, it was already there, indicating 6 stops from middle grey to blown out. (an arbitrary aside, I noted that the green indication of middle grey was measuring at about 42 IRE) Then I turned down my light source until the white chip was showing as green (42 IRE). Then I close down the aperture until the black chip turns purple (no exposure). I had to close it down 2 stops for this to happen. Then I add it all up: 6 stops from middle grey to blown out, plus 8 stops from middle grey to zero exposure, equals 14 stops.

What am I missing? Has anyone else tried to measure this yet?

Also does anyone know what ISO level has been recommended by Red as the base ISO? I can't seem to find that information anywhere. I tried this test at 800 and 2000 ISO, both adding up to the same number of stops.

haven't done this test yet…so let me know which one works?? thanks

 

[Robert Ruffo New]

John is correct. Here's how we measured it:

Shoot a high dynamic range scene, like out a window in a dim room, seeing to a bright sunny day outside, with some objects near the window.

Go through all the stops on your lens, noting each f top .

When you get to the smallest f-stop, add ND in steps, ND 0.3, 0.6. etc….

Now playback, adjusting RCX in ISO so that all moments on your clip match in brightness, and render out a sample of each clip piece correctly adjusted.

If your scene is very bright, you may want to add the first steps of ND (say nothing, then 0.3 and 0.6) when the lens is wide open, and run all other stops at 0.6, then just add heavier ND to the end when your lens is fully closed. You can also double-up the ND at the dark end to add even more stops there.

As you compare the clips, you'll learn a lot, and by tracking a single object or area of the image, you can decide for yourself how much room you have for over and under exposure, based on the levels of noise (at the low end) or loss of color rendition (at the high end) you find acceptable

 

[dio Zafi]

John is correct. Here's how we measured it:

Shoot a high dynamic range scene, like out a window in a dim room, seeing to a bright sunny day outside, with some objects near the window.

Go through all the stops on your lens, noting each f top .

When you get to the smallest f-stop, add ND in steps, ND 0.3, 0.6. etc….

Now playback, adjusting RCX in ISO so that all moments on your clip match in brightness, and render out a sample of each clip piece correctly adjusted.

If your scene is very bright, you may want to add the first steps of ND (say nothing, then 0.3 and 0.6) when the lens is wide open, and run all other stops at 0.6, then just add heavier ND to the end when your lens is fully closed. You can also double-up the ND at the dark end to add even more stops there.

As you compare the clips, you'll learn a lot, and by tracking a single object or area of the image, you can decide for yourself how much room you have for over and under exposure, based on the levels of noise (at the low end) or loss of color rendition (at the high end) you find acceptable

nice..thanks!

 

[Adam W Watson]

SOLVED! Thanks John for the heads up about the "exposure" indicator. This time I went strictly off the waveform and this time only looked at one point instead of two. What I did this time was pick a bar on the chart. Then I opened the lens wide and lit the bar so that it just barely blew out completely on the waveform. Then I closed down the lens all the way (6 stops) and looked at where the bar measured on the waveform (55 IRE) and remembered that. Then I turned the light down as much as I could in my situation (not totally dark) and set the aperture so that the bar once again read at 55 IRE. Then I closed down the lens all the way (3.5 stops). From there I couldn't control the lighting anymore so I switched to the shutter. I had it started at 360° to begin with for this test. I reduced the exposure further by halving the shutter each stop. 180°, 90°, 45°, 23° ect until the bar disappeared on the monitor and the waveform. I added the first 6 stops, then the 3.5, then the rest and came up with 17 stops. Thanks everyone.

But I'm still curious as to what the base ISO is meant to be, I know its a little subjective but I'm surprised that a recommendation hasn't been given by Red. Anyone know?

 

[Phil Holland]

But I'm still curious as to what the base ISO is meant to be, I know its a little subjective but I'm surprised that a recommendation hasn't been given by Red. Anyone know?

Unofficially I'd say ISO 800, however since you have access to the total dynamic range at all ISO values safely between ISO 250-2000 it's more about where you rate your midtones and how you choose to expose for a given scene. You could make an argument for 1000/1250 as well. The base sensitivity of what the Dragon sensor can see has slightly increased over Mysterium-X.

There is a chart balanced ISO value however that will give you both an equal amount of values on either side of the midtones. The chart test should reveal what that number is

 

[Martin Stevens]

So..... Phil and John, can you say at what ISO on the Dragon you believe you get 8 stops over and 8 stops under.

ISO 2400 or higher perhaps?

 

[Adam W Watson]

Not finished with my tests, but wouldn't go over 800 unless I had to, it introduces too much noise on the low end. Not sure what the stop balance is yet though.

 

[George A.]

Purple is not zero- it's the beginning of the last 10% - and that last bit on dragon holds a lot of magic you're writing off.

i've measured the DR practically, against a meter, with a xyla -21 chart, and with my sekonic calibration chart. All come out at over 16. ISO makes no odds in the final outcome.

Hi John,

Is there any penalty in shooting ISO 250 vs ISO 800 for instance? Are you clipping "sooner" when shooting ISO 250?
What's your experience?

 

[Adam W Watson]

I haven't tested the ISO in relation to the floor and ceiling of range and where the best balance lies in that, but I did test the ISOs for noise and got several people to give a "blind taste test" for noise/ISO. The general consensus was that The noise levels became detrimental past 800 and that it was the best balance at 640. Perhaps this is why Red isn't touting an ISO like they did when the MX came out, since there seems to be no improvement in that department.

However The Redcode, like the DR, is greatly improved. I had a hard time telling much of a loss in detail in gradient portions of the image even all the way up at 18:1. This is great considering I wouldn't recommend going higher than 9:1 on the Epic MX.

The flare is strange. Using a Master Prime pointed at the sun at f/22, there was a strange array of dot flares in a grid pattern. It looks cool on one hand, but maybe not what some people will want. However if you open up the iris some the dots expand and it stops being distracting.

 

[Adrian T.]

I had a hard time telling much of a loss in detail in gradient portions of the image even all the way up at 18:1.

Gradient portions are easiest to compress using wavelets. Small high frequency details are hardest. So be careful shooting for example foliage or lawns with a high compression rate.

The flare is strange. Using a Master Prime pointed at the sun at f/22, there was a strange array of dot flares in a grid pattern. It looks cool on one hand, but maybe not what some people will want. However if you open up the iris some the dots expand and it stops being distracting.

There will be a new OLPF. Read this: http://www.reduser.net/forum/showthread.php?111400-Dragon-Flare&p=1320952&viewfull=1#post1320952

 

[Adam W Watson]

Ah well that's good. Too bad we have to send them back out for the upgrade when we just got them in the first place last week though.

 

[Evin Grant]

Adam, how are you viewing/presenting the EI noise to people? I've found that it looks much noisier viewed out of RCX at lower resolutions than if you do a full debayer output.
The 2K Ei2000 files I've projected looked extremely good to me, granted they have a nice granular texture to the shadows so not strictly speaking "clean" but no worse than any 500 ISO film stock I've seen projected. Also make sure you have the NR set to 6 in RCX if you're shooting above EI 1000. Lastly there we're some shadow point changes (cleaner) in build 5.1.35 so if you haven't upgrade yet I recommend you do and re-test.

I'll upload a sample QT soon.

 

[Adam W Watson]

Thanks, You're referencing the Denoise function yes? It goes up to 6 so I assume that's the only function you're referencing. I am at 5.1.35. It's evident in the monitor when magnified but to show people I did a full resolution screen capture of each exposure/ISO from RCX and had them flip through the pictures zoomed into a portion of the screen at 100% pixel-to-pixel. I'm not complaining, 800 is plenty sensitive.

 

[Evin Grant]

The monitor output on the camera is a very quick debayer and not indicative of the actual noise upon output. The still frames are more accurate but IMHO you need to see the noise in motion because not all noise is created equal, the more organic it is the more you can generally tolerate in the image before it become distracting. The Dragon has the most organic noise texture of any digital camera so far.

Here is a full debayer 1080P Prores 444 QT shot at EI2000, exposure was set using a Sekonic L308 meter and the histogram, shot at 28mm T2.1 23.98FPS, RC6 180º 2800K -5 RC3/RG3
http://www.evingrant.com/flicks/Bear2000.mov

Now i'm not saying this is noiseless, but when I project it on an 8' screen it looks quite cinematic and the shadow texture is not distracting to me at all.

 

[Martin Stevens]

Evin,

Your 2000 ISO dog footage looks noiseless to me, even though I'm sure there is technically
noise there. Film at ISO 500 has far more noise/grain.

 

[Phil Holland]

On further examination of this clip, Bear2000, I have found myself writing science fiction based on "noiseless and mechanical dogs" late into the night