The toe and the shoulder Dragon DR

[David Battistella]

So,

I've be been looking at a lot of dragon footage and really digging into some grading on the new OLPF. I have a bit of a theory on why people are getting noise or not.

There's a couple of important considerations.

-You can not treat Dragon just like MX.

Don't fall into the trap of thinking you know the camera because the camera looks the same. It's not. It's a different camera. Pretend you never used a RED before.

if you've owned or used an MX a lot in the past few years, you need to start unlearning your exposure techniques a bit. I'm speaking very specifically about highlights. The shoulder was a very hard clip on the MX and that is gone on Dragon. This means you can open up without worrying so much about clip.

If you keep exposing like you would on MX you will underexpose. The MX hard clipped, so the images favored slight under exposure then digging out of the shadows.

its sorta the opposite on dragon. You should be thinking about filling up the histogram and "putting back the blacks". This is why maybe it seems more noisy, because we are used to digging stuff out of the shadows and protecting highlights, but this exposure strategy could bite you a bit on dragon if you are not careful.

You can hit it with more light (like some people are reporting to treat it like 35 and expose it one stop over). This is where the 2000 is the new 800 comes from. If you light dragon, or more like, if you hit the sensor with enough light, pushing to 2000, 3200, 6400 is doable because it's a much cleaner sensor.

Ill add more here later, but really two fundamentals.

1. Don't treat it like MX
2. Think in terms of putting back the black instead of trying to dig stuff out of the blacks.

Battistella

 

[Mark Pugh]

So,


You can hit it with more light. This is where the 2000 is the new 800 comes from. If you light dragon, or more like, if you hit the sensor with enough light, pushing to 2000, 3200, 6400 is doable because it's a much cleaner sensor.


Battistella

No, David, that quote from Jim came because you could hit Dragon (with OLPF1) with LESS light and have equivalent image noise to MX.
Also, with OLPF1 you don't need to crush the blacks to tidy up noise. I routinely haver a curve in camera, lifting the shadows with OLPF1, in camera, and it's clean at 800ISO.

 

[David Battistella]

I'm talking highlight OLPF strategy. I should have prefaced with that. It sounds like old OLPF would employ an MX style strategy protecting highlights a bit more.

in anticipation of interchangeable OLPF's it would be great to share lowlight OLPF strategies as well.

Battistella

 

[Christof Paul]

So,


its the opposite on dragon. You should be thinking about filling up the histogram and "putting back the blacks". This is why it seems more noisy. Because we are used to digging stuff out of the shadows and protecting highlights, but this exposure strategy will bit you a bit on dragon if you are not careful.


Battistella

i can't shoot my Epic MX to protect highlights and dig out stuff in the shadows. I get too much noise with MX if i try and do that. Does that mean that i have to prepare to protect the shadows even more with Dragon?

 

[David Battistella]

i can't shoot my Epic MX to protect highlights and dig out stuff in the shadows. I get too much noise with MX if i try and do that. Does that mean that i have to prepare to protect the shadows even more with Dragon?

No, that's not what I meant. I mean more that if you underexpose dragon and try to recover (like if you underexpose a bit to protect highlights) then deploying that same strategy (protecting highlights by underexposing) is not as necessary as it was on MX. The Dragon rolls off nicer in the highlights (closer to film), it still clips, but the clipping point is about a stop and a half higher. So if you protect highlights (too much) on Dragon (if you expose it like MX) then you will end up underexposed and digging stuff out of that underexposed part of the image is where you will feel the image has more texture.

My overall point is to make sure you really take this thing out for a spin because its a new camera and some of the "old way" of doing things specifically on MX to get certain results, won't yield the same results.

Test.

Battistella

 

[Meryem Ersoz]

I just finished 2-1/2 weeks of shooting a lot of lowlight set pieces. I basically told the gaffer to light a stop over whatever we metered for, and everything looked great when we started pushing sliders around in RedCine - basically affirming what David is saying, don't worry about clipping or over-lighting. If you hit the sensor with more light, you get a more flexible image. It is preferable to reduce exposure than it is to push exposure.

RAW works best when you understand it as subtractive medium, not an additive one. It's better to overlight for a fatter histogram and then subtract data if needed, rather than have a weak histogram and expect to push it.

The one thing that hasn't changed between these sensors is the rule that fat histogram = flexible image. A weak histogram limits the flexibility of your images. I can't stress that enough.

And sometimes you have to fight that fight pretty hard, when the DP or the director assesses their images on uncalibrated rental monitors. You can't light lowlight scenes effectively by eyeballing them on a monitor, without confirming the histogram. Monitors can lie to directors and DPs, but the histograms and false color (in combination) won't....

 

[Mark Pugh]

Doubling the amount of light to make up for the limitations that OLPF2 imposes on the sensor. Fantastic...
The decision to switch to this OLPF might help sink this camera. That would be a shame. I hope I'm wrong.

 

[Ryan Emanuel]

I don't think you have to double the light, you can expose normally for 250-800 ISO and be fine with DEB and a gamma curve. You can expose normally at 2000 ISO and be fine as long as you fill blacks that you want detail for and crush blacks that you don't.

 

[Christof Paul]

No, that's not what I meant. I mean more that if you underexpose dragon and try to recover (like if you underexpose a bit to protect highlights) then deploying that same strategy (protecting highlights by underexposing) is not as necessary as it was on MX. The Dragon rolls off nicer in the highlights (closer to film), it still clips, but the clipping point is about a stop and a half higher. So if you protect highlights (too much) on Dragon (if you expose it like MX) then you will end up underexposed and digging stuff out of that underexposed part of the image is where you will feel the image has more texture.

My overall point is to make sure you really take this thing out for a spin because its a new camera and some of the "old way" of doing things specifically on MX to get certain results, won't yield the same results.

Test.

Battistella

i never underexpose to protect the highlights with MX. i prefer clean shadows to "not clipped highlights". does that mean if i expose dragon and mx side by side and protect the shadows with both the same way, i will get 1.5 stops more detail (less clipping) in highlights? i wish someone could do this and post raws, sound pretty simple to me.
although: isn't dragon supposed to be 16.5 stops DR? my epic mx is nowhere near 15 stops. that number would mean at least 1.5 stops more detail in shadows AND 1.5 stops less clipped highlights?
is there a side by side available with raw files? the camera has been out long enough now! is there one on red.com?

 

[Timur Civan]

Treat er like 35. shoot 800-1000 Stop over and pull it. Or set the camera to 500 and call it a day. About the same result.

 

[Jan Reiff]

i never underexpose to protect the highlights with MX. i prefer clean shadows to "not clipped highlights". does that mean if i expose dragon and mx side by side and protect the shadows with both the same way, i will get 1.5 stops more detail (less clipping) in highlights? i wish someone could do this and post raws, sound pretty simple to me.
although: isn't dragon supposed to be 16.5 stops DR? my epic mx is nowhere near 15 stops. that number would mean at least 1.5 stops more detail in shadows AND 1.5 stops less clipped highlights?
is there a side by side available with raw files? the camera has been out long enough now! is there one on red.com?

you prefer clipped highlights with the MX ? really ?

 

[Christof Paul]

you prefer clipped highlights with the MX ? really ?

if i have to choose between clipped highlights and noisy or muddy shadows, i will definitely protect my shadows and make sure no detail is lost there and let the highlights clip. in my opinion crushed blacks almost always look digital and even if they did not, i just don't like that contrasty look. of course that is a matter of personal choice.
but in terms of shooting an extreme backlit scene i have no problem letting the highlights go to make sure the backlit person's (dark) hair does not turn into a muddy blotchy mess and i definitely would not want to crush it either or use noise reduction or some other post processing trick, if the detail is lost, it is lost. i find that adding grain overall gives even clipped highlights a natural analog feel.
having extra stops on top or bottom is always helpful regardless of course.

 

[Gunleik Groven]

Treat er like 35. shoot 800-1000 Stop over and pull it. Or set the camera to 500 and call it a day. About the same result.

Smart guy.

It's actually not very hard to shoot the Dragon... Funny that the more DR we get the more we complain...

 

[Björn Benckert]

To me this is all very confusing. I mean the dragon or any digital sensor clips just as hard as the MX. Since the dragon sensor is just as linear as the mx, however the way the colours are represented trough the r3d developing a lot of that linearity is used for dynamic rounding in the highlights for the olpf2. The olpf2 also eats a lot more light so simply it needs more light then the olpf1.

So olpf2 is like ND.3 and no red blob flairs around highlights.
olpf1 is like taking that ND out + getting the red blobs in.

What I find odd is that if I look at a dragon frame in rcxp and set it to linear and camera RGB then I get a black frame. I guess thats a bug but I find it od that we can no get a linear clean view of the source.


Another thing, when I raise the iso the histogram gets squished, this I find to be wrong as it is not linear lift. To me thats mathematically wrong, the picture should if it's 50% exposed at 800iso get a uniform lift to hit 100% exposure when iso is set to 1600iso, this is not what's happening now instead when doubling or increasing iso the lowlights climb more than the highlights. I assume thats an attempt to make r3d's to behave like film but to me thats a gamma lift it's not a clean increase of iso.

So basically a picture shot att 400 iso with double the amount of light as the same picture shot at 800 iso but with half the light should look mathematically identical. Well thats not the case as the histogram is not getting a uniform lift but instead clamped towards the top.

So I'm guessing these sliders behave under a dynamic lut that keeps the image from clipping in a linear way. To me all such thinga are grading... sliders like flut and iso should not behave like such they should be linear, no?

Sorry to derail a bit.

 

[David Battistella]

Bjorn,

Isn't that just what "others" do in camera, just that this is being done in post software? It's just how the matrix is mapped. Have you noticed this with the MX r3d as well? or just Dragon?

Battistella

 

[David Battistella]

Smart guy.

It's actually not very hard to shoot the Dragon... Funny that the more DR we get the more we complain...

Plus 1. I tried to quote Timur's post as well, but instead I hit the wrong button.

Battistella

 

[Elsie N]

Doubling the amount of light to make up for the limitations that OLPF2 imposes on the sensor. Fantastic...
The decision to switch to this OLPF might help sink this camera. That would be a shame. I hope I'm wrong.

You are '-)

 

[Gunleik Groven]

Björn, you're right.
At some point RLF became the "reference" LUT from which they're all derived. That is generally OK, but I do see the issues popping up for VFX work.
Graeme/RED will advocate "libearizing RLF with a cineon to lin LUT" in f. ex. Nuke, but for a multitude of reasons which you hint at, it would be good if the Linear was actually just that, and didn't have a knee in the top. Att David, that's really not a big issue to handle in the situation where linearity is preferable.

I haven't really tested this, but I have noted the behaviour Björn describes, and they're confusing.
On the MX lin@320 ISO was allways a "safebet" even with this behaviour, as that seemed to be the calibrationpoint. I haven't had the need to test that for Dragon, but your point is pretty valid for a few of the workflows I have been involved in.
Maybe Mark could chime in?

 

[Mike Garrick]

So where does this leave the 250-320 sweet spot........ or has it moved ?

P.S. This may appear a dumb question in such learned circles but I've a shoot coming up with it and I've had the camera for 24 hours.

 

[David Battistella]

Its curious and I see both sides, wanting a straight mathematical representation and wanting to adjust the curve as you move around th scale. Maybe there needs to be an expert version of RCX.

battistella