My "Field" latitude test

[Jeremy Neish]

We are prepping to film a feature movie sequel on Red. My camera #63 will be doing the majority of the work. The D.P., Marvin Rush, is very saavy with digital cinema having worked on hundreds of Star Trek episodes many of them digitally. He wanted to know the real world limits of the camera, especially as far as latitude, so we took the camera out to one of the locations and shot some tests. These tests are by no means scientific / lab tests, but I do feel they are fairly accurate real world tests.

Here are my first results. The process was as follows:
- Shoot a range of exposures, separating each by 1 stop, color temp ~5600 (mid day sun was the only light source)
- Save 4K Tiffs from Red Alert (Rec 709, Denoise Max, Debayer Detail High, OLPF Compensation Off, ISO 320, everything else Defaults)
- Import into AE (Project set to 32bit float)
- Start in the middle (5 stops) and work in each direction until:
- Image completely underexposed in noise
- Lightest chips clipped

Basically I tried to make each of the images look as similar in exposure as possible, I used a combination of the "Exposure" and "Levels" filters to match the the darkest chip, the middle chip and lightest chip between each image. Exposure would get me in the range then Levels to adjust for linearity changes.

http://redhelp.net/wiki/index.php?title=Latitude

If anybody wants to mess with my results, I'm uploading all of the source TIFF files and the AE file to Rapidshare. As soon as it's done uploading I'll update this message.

Updated with color temp information.

 

[Robert Aldrich]

So what, 9 stops before it begins to look a little grainy? Can it be tweaked (in editing) beyond that?

 

[Fredrik Callinggard]

So what, 9 stops before it begins to look a little grainy? Can it be tweaked (in editing) beyond that?

Unfortunately not - it is already tweaked, this is when the information starts to breakdown.

(My preference is to shoot with 7.5-8 stops in mind before it breaks - just as this test shows. This is not the end of the world. I love my cam and the images but it's good to know it's limits)

Fredrik Callinggard

 

[Hans von Sonntag]

Personally I prefer real life scenes and not charts. In real life I find 8-9 Stops at 250 ASA the range of the RED camera. As said many times before it depends how much noise one excepts. I also have no big fear for overexposing and do not follw the "protect you highlights" dogma too much. One can achieve a nice roll-off later if footage is correctly debayered.

Thanks for the test anyway, it shows how destracting noise can become.

Hans

 

[Chris Nuzzaco]

What color temperature did you shoot under? Thats a very important thing to note.

Thanks!

 

[Jeremy Neish]

Good point. It was mid-day, only light being the sun. So ~5600. I'll update the site with that info.

 

[Matthew Bennett]

Such strange noise...

 

[Adam Levins]

8 stops is plenty. There is lots of talk about the RED having less latitude than film but one advantage is that you can see what you are getting on the day.

lighting packages will needs to be up to par on any ambitious production. The Challenge I feel, is learning how to adapt the silks and lighting package to this new camera. Having said that it is quite easy if you have a tent as you can see the results on set!

 

[Fredrik Callinggard]

8 stops is plenty. There is lots of talk about the RED having less latitude than film but one advantage is that you can see what you are getting on the day.

lighting packages will needs to be up to par on any ambitious production. The Challenge I feel, is learning how to adapt the silks and lighting package to this new camera. Having said that it is quite easy if you have a tent as you can see the results on set!

Hmmmm well there's a reason why people prefer to shoot negative film instead of reversal.

Pick up your old 35mm still camera and shot of a few shots with some negative film and then shot the same scenario with positive film. Then you will understand the reason for wanting more stops.

The problem is that you have to crunch the blacks so quickly to not get undesirable noise. What if you don't want it to be high contrast? 2-3 stops in shadows is not that much to play with (as goes for highlights as well).
Of course it's possible to live with it but it does give you restrictions that sometimes lets you down.

I know that RED is on the case for this and that the blue channel noise will improve. As for know I have to wait a little with my "prince of darkness" shoots.

Fredrik Callinggard

 

[Andrew M.]

Very good test!!

Stop 0 and one, are you losing blue there?
You start clipping there on one color and two other colors are still good. Left lower corner of the color chart, at stop 3 it is almost blue at stop 1 and 0 it is pinkish.

You can push two other colors extra stop there and recover from clipping.

Did you use any noise reduction in stop 9?

Also I see you are counting from zero "0"
I am not the expert but if this is the standard to count stops from zero and bits from one then Mysterium is only 11 bit sensor, counting from 0 to 11.

For me the stop 9 in your test is stop 10 in mine.
I count from one, since Mysterium bits I count from one to 12

Holly $%/t! what a clash of two worlds, digi and film, analog and numeric.
This forum is the best for convergence technology in moving pictures production!

 

[Jeremy Neish]

Very good test!!

Stop 0 and one, are you losing blue there?
You start clipping there on one color and two other colors are still good. Left lower corner of the color chart, at stop 3 it is almost blue at stop 1 and 0 it is pinkish.

You can push two other colors extra stop there and recover from clipping.

Did you use any noise reduction in stop 9?

Also I see you are counting from zero "0"
I am not the expert but if this is the standard to count stops from zero and bits from one then Mysterium is only 11 bit sensor, counting from 0 to 11.

For me the stop 9 in your test is stop 10 in mine.
I count from one, since Mysterium bits I count from one to 12

Holly $%/t! what a clash of two worlds, digi and film, analog and numeric.
This forum is the best for convergence technology in moving pictures production!

I noticed the pinkish tint on the most overexposed, I wasn't watching the camera's display at the time, but I suspect the blue channel was clipping. I tried compensating, but couldn't quite nail a neutral grey.

No post production noise reduction was added, other than in REDAlert. And it was set the same on all the images.

You are probably right, I should have started counting at 1, this would have made the first blush latitude seem one stop wider. And I thought about it after I was done, but then I had second thoughts since 0 was obviously clipping, it kind of made sense to count #1 as the first stop without any clipping.

 

[jimhare]

So how much of this can be seen on the histogram while shooting, and how much needs to be old fashioned knowledge and experience?

I think I'm keeping up with the science okay, I just want to start predicting when to call this into action in the real world!

Thanks for doing this,

Jim

 

[Chris Nuzzaco]

I know that RED is on the case for this and that the blue channel noise will improve. As for know I have to wait a little with my "prince of darkness" shoots.

I hate to break it to you, but blue channel noise, especially under tungsten light sources, is a physics issue with the sensor design. There's no real fixing that unless you optically compensate via filters (ie. 80A and hot mirror to kill IR pollution).

I noticed the pinkish tint on the most overexposed, I wasn't watching the camera's display at the time, but I suspect the blue channel was clipping. I tried compensating, but couldn't quite nail a neutral grey.

A pinkish/magenta tinged image is the result of green channel clipping. I'm really not too surprised by that however, green is very easy to capture, and under all color temperatures, seems to be very plentiful.

 

[I Bloom]

I hate to break it to you, but blue channel noise, especially under tungsten light sources, is a physics issue with the sensor design. There's no real fixing that unless you optically compensate via filters (ie. 80A and hot mirror to kill IR pollution).

Not sure that is true in that it may be "fixable" in the sense that DSP and compression may be tweaked to compensate for the way it "looks". Blue will always be at a disadvantage, but I'm excited to get my hands on 16.

 

[Andrew M.]

I don`t understand one thing in DR calculation in the film domain.
On the picture marked 'Stop 2' I see the white vertical bar (on the left chart) at near maximum value or very near of 2000 I don`t have TIFF files yet to be more precise.
I am guessing since on the previous picture `Stop 1` the right most white vertical bar (left chart) is almost at the clipping value of 4095.


At the same time on the same picture I see the black bar at the low value, there is 8 bars so 8 stops minus 2 since it is on the picture 'Stop 2' = (- 10 stops)

I know that you were stopping down aperture for all next pictures but right on this picture 'Stop2' you were going down in DR to almost full 10 bit range and there is no noise in the blacks there, where on the picture marked Stop 10 there is visible noise in all blacks on the picture?

Is this possible that in post you compressed histograms a bit?



On the Mysterium when all bits are zero then black is most black, no signal at all.

(actually I think black is all bits 1111 1111 1111 so 4095) forget this one though, just for gigs.

Again, on the Mysterium when all bits are zero then black is most black, no signal at all. (AE Color Finesse)

0000 0000 0000 0000 = 0 Stop 0

So first bit up 0000 0000 0001 = 1 will be noise -Stop 1

one stop up 0000 0000 0010 = 2 almost no noise Stop 2

again one stop up 0000 0000 0100 = 4 no noise Stop 3

0000 0000 1000 = 8 - stop 4
0000 0001 0000 = 16 - stop 5
0000 0010 0000 = 32 - stop 6
0000 0100 0000 = 64 - stop 7
0000 1000 0000 = 128 - stop 8
0001 0000 0000 = 256 - stop 9
0010 0000 0000 = 512 - stop 10
0100 0000 0000 = 1024 - stop 11
1000 0000 0000 = 2048 - stop 12
1111 1111 1111 = 4095 - stop 13

So if you accept a bit of noise you get 11 stops and completely smooth surface 10 stops or I am blind.

I think it would be the best to do the same test on film and compare.

 

[Jeremy Neish]

A pinkish/magenta tinged image is the result of green channel clipping. I'm really not too surprised by that however, green is very easy to capture, and under all color temperatures, seems to be very plentiful.

You are correct, I was thinking backwards.

 

[Jeremy Neish]

Is this possible that in post you compressed histograms a bit?

Yes, as I mentioned in the notes, my goal this "real world" test was to match the values of the brightest bar, a middle bar and the darkest bar between all of the photos. I used the exposure and levels filters to do so. So yes, there were times when just adjusting the exposure didn't cause a linear match, so I had to use levels to snap them back into position. This effectively compressed the histograms.

 

[Chris Nuzzaco]

Not sure that is true in that it may be "fixable" in the sense that DSP and compression may be tweaked to compensate for the way it "looks". Blue will always be at a disadvantage, but I'm excited to get my hands on 16.

What I mean is that blue isn't really ever going to look very good under tungsten unless you filter the camera for its native balance (5000K according to the Red Manual). If you enter the digital domain with a poor physical exposure, you'll never have anything better than that in my opinion. Its best to sample a healthy exposure. I have a few other theories regarding the Red ISO, and image noise, but I won't post about it till I conduct a test (later this month in fact).

No channel is ever really at a disadvantage though, you can get a clean blue channel under tungsten without filtering, but it requires blowing out the red and green channels because the physical exposure gap between them is pretty big, several stops in fact. If you want a healthy exposure in all three channels without blowing out any other channels, then yes, blue will always be at a disadvantage under tungsten. Just some food for thought though... If the sensor was balanced for tungsten, we'd have people talking about red channel noise when shooting under daylight without optical correction.

I own a Viper like camera system where there is no attempted white balance (thus the images have a strong green cast all the time), and when I shoot unfiltered under daylight, I have red noise issues, and blue noise issues under tungsten. In each case however, I can easily expose each channel properly if I want, but I need to sacrifice the other two channels exposures in some way. Thus the need for optical filtering.

I once did an experiment and posted it somewhere at the Reel Steam forum where I shot a still life scene under tungsten, making three different exposures, one optimized for each channel, then in post I combined the "healthy" RGB channels and created a virtually noise free balanced image. Each channel was pristine looking, even blue, but blue required an exposure roughly two times greater than green!

Hope that helps explain what I mean

 

[Andrew M.]

I made simple experiment shifting blue color up (more exposure by 1.5 stops) then red color unchanged and green underexposed 1.5 stop.

Then I treated each channel as a black and white except the part that overlapped.

I got extra 3 stops out of these maneuvers but now I am trying to figure out how to bring the color balance in the underexposed area so it looks natural. Overexposed I figured out by playing in Color Finesse.

 

[Chris Nuzzaco]

Have those .r3d files uploaded yet? I'm eager to play with them.