13.5

[Rich Schaefer]

So, I can confidently say I am getting 13.5 stops out of my Red MX camera.

I posted this in mid February. I did this test for my own knowledge before taking the camera out on a commercial shoot. I wondered about my results because I changed the stops with the lens iris and camera shutter.

Just got #115 back. MX is bad ass! In my testing today I pulled 13.5 stops of DR from my camera!

I did the stops with a lens. I will do it again with lighting and a meter not with stops.

Then Jim posted his early Arri DR test that seemed to agree. (There is some concern about calibration.)

It appears that the M-X sensor sits at 13.5 stops on their test at 1st blush.

Jim

This week I re-shot my test and instead of using the lens to accomplish my stop differences I used light (and shutter). To be specific I used 3 light meters to measure it in footcandles. My findings were again 13.5 stops. Here is the specific area of the chart I am looking at to determine if the camera can see a delineation between the two shades.

I did TIFF still frames of the exposures. The Red MX was at 500iso (really just metadata) Here is +4 stops:

Here is -9 stops and -9.5 (the bar to the right is ½ stop darker.)

I stand behind my 13.5 stops! BUT, here is the real zinger. When I look at -10 stops under I cannot see the difference of the shades. However, when I play the -10 stops at speed with my MacPro and RedRocket card I can then see the delineation of the two shades. Granted it is noisy and I am pulling up the image hard with RedCineX but I can make out the line. Here is -10. You decide, you do the math:

Sorry, its so noisy. Its better when I roll the footage.

Like I said, I definitely stand behind my 13.5.

I am not enough of a post guru to argue 12 bits = 12 stops…. I am a shooter of 22 years and I know lighting and cameras and I know what I am seeing. The Red MX is seeing a delineation at these levels. Perhaps Graeme can contribute here to the bits=stops theory.

To the doubters, I say, test it yourself and lets look at your findings.

To the other camera manufacturers, I say, Give me your cameras to test and I will wring them out just the same.

All the Best,
Rich

 

[Jarred Land]

Thanks for doing this test Richard... what light source was used?

 

[Dan Hudgins]

Downsample to check DR without so much grain showing

Downsample to check DR without so much grain showing

I am not enough of a post guru to argue 12 bits = 12 stops…. I am a shooter of 22 years and I know lighting and cameras and I know what I am seeing. The Red MX is seeing a delineation at these levels. Perhaps Graeme can contribute here to the bits=stops theory.

...

Rich

You can get a better reading of the DR if you downsample the images so that the "noise" dithered image data is averaged using all the pixels.

If you make it about 240 pixels wide (the whole image) and watch that uncompressed at speed, then you can see better and use a curve to expand the shdow detail after the downsize.

You can get more that 12 stops from 12 bit ADC since the "noise" has dithered the signal past the LSB threshold. Normally the fixed and moving pattern noise is higher than that, so they must have improved both? Before I (and others that posted) was seeing some vertical fixed pattern noise it seems, if you shoot 10 seconds of black with the lens capped in total darkness then sum those frames (the de-Bayer program I am working on will do that for TIF as well) then you can see the fixed pattern noise. If you then subtract the fixed pattern noise you can see the moving pattern noise when you process the same 10 seconds of black with its average subtracted.

I put that in my de-Bayer for both raw data and TIF so RED footage could be processed as well, but I have not gotten any RED dark field frames yet, so if someone can shoot an R3D for me of total darkness it would help with that development. Dark field frame sets for both M-X and M sensors would be a help as well so I can see if it helps one or the other or both... more than 10 seconds each is even better since the more frames you average the less moving pattern noise you get so the subtraction is cleaner. To see effects on footage I would also need some somewhat underexposed frames with subject to process the fixed pattern noise from.

Since you can see the lower stops when the footage plays, if you average the whole shot with the chart, say 65535 frames, then you can average out the noise and see if there is any real data there, its sort of an HDR result, but it shows what the individual frames are holding under the "noise".

In a real image with better than 9 bits in the RGB after monitor gamma correction you should not be able to see ANY noise in the image. So if you look at the "linear" RAW image no noise should be visable if the exposure is right, if you see noise then the data is not clean to the 9th bit.

 

[Rich Schaefer]

I actually used a 10k. I did so to get the levels. It's hard to light a single subject to 14/15 stops of latitude..

So, indeed the mx dose 13.5 stops with tungsten light. Pretty good for a daylight sensor.

Best,
rich

 

[Deanan]

I am not enough of a post guru to argue 12 bits = 12 stops….

There are ways of getting more than 12 stops out of 12bits at the sensor level.
12=12 is only true if you have a very strictly linear sensor. (pixel, a/d, etc)

 

[Rich Schaefer]

Thank you Dan for your thoughts. I would love to have you do your thing and to get you oppinion.