Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Exposure Placement

I

Irri

Banned
Joined
Mar 29, 2007
Messages
499
Reaction score
0
Points
0
I've been having an argument, well more of a discussion with a colourist via DoPs about RED exposure. He's a nice guy and good colourist, but we disagree on something...

He says it is much easier to work with stuff that is underexposed. Now, I don't think he really knows the RED process, so here is my question:

Say you shoot something very flat, for example only using 1/3 of the sensor's dynamic range, does it matter where you place that 1/3? I could put it at the bottom 1/3 of its range, the middle, the top etc. I could then move that 1/3 wherever I wanted in RED Cine/Alert, but does it make a difference where it was placed in the sensor's range when shooting?
 
Here is the technical thing I got explained for me:
In the video world the brightness and light is messured in watts. And the visible part of the image is messured from 0,3 W to 1,0 W. But when your making an image brighter in post your pulling brightness up from under the 0.3 W. So you still got information from the brightness range that you can't see.
Do you understand?
I'm a little drunk and speak kinda bad english right now :)
 
yup, it makes a difference.

you have higher precission the "lighter" the image is.

Actually the precission doubles with each bit.

But it also depends a bit on which format and how you post.

My 2c

(if I may) Have a look here:

http://www.reduser.net/forum/showthread.php?t=12609

You can always have them blacks back... That's not the hard part.
 
For a low dynamic range scene you ETTR - or expose to the right, ie, as bright as possible without clipping. For a scene with greater dynamic range than you record, or where shadow detail is more important than highlight detail, you have to increase the exposure somewhat compared to that highlight clip point.

Graeme
 
I'm on the "expose to the right" group (placing the 1/3 in the highs). My thought was that if you have to pull something down, you will have cleaner shadows. Also, there's more data (precision) in the highs as the others have already mentioned. I remember seeing a greyscale chart someone made that represented what the digital cmos sensor sees and there were a lot more bright to mid values then there were darks. Here's something about exposing to the right:
http://www.luminous-landscape.com/tutorials/expose-right.shtml

However, I recently spoke with Michael Cioni of Plastercity at NAB and he has the opposite opinion. Because I respect his knowledge, it did cast some doubt and I've resigned myself to waiting for my own testing.

Edit: But then again, I highly respect Graeme's knowledge, so I'm content with being in the ETTR camp.

As far as my experience with my Canon 40D, exposing to the right is great. The only problem is that my LCD isn't a good judge of when I'm clipping. But I think that now that we can view RAW for what it is and there are monitoring safe-guards (false color), it's easier and better with the Red.
 
  • Thread starter
  • Banned
  • #7
So my theoretical scene which only uses 1/3 of the dynamic range, and could be placed at many different places in the sensor's range without getting close to the noise floor or close to clipping, should be placed at the upper end (or right)?

I guess that is the opposite to what the colourist is thinking and also different to what I was thinking (that it made no difference).

Is this based on the idea that of a 12 bit (4096) image, the last stop is 2048 of the whole image, or am I getting confused?

edit
I just wanted to add that in this hypothetical shot there is no clipping. We are moving something like 3 stops in 9 stops of available dynamic rage.
 
Edit: IRRI, we have 11.3 remember! :)

As far as I understand (and from most I've seen) you are right on...

Then we only need the exposure meters to be inserted before metadata so we know when the sensor actually clips.
You're still not home safe, though:
Problems arise when you trunctuate the info before CC....
 
On a side thought about metadata affecting exposure... does white balance affect it? Or, in other words, can I forego white balance altogether and leave that for post (I don't have to appease any clients on set); or do I have to set white balance to get a proper exposure? I thought I didn't, but then someone on the forum said it affect it.
 
Metadata doesn't (AFAIK) affect exposure, but the metering of the exposure...

To your question
Depends on your workflow.

If you go through RC/RA and treat proxies as proxies (that is, not finishing even from Compressor batch-created proxies) you can leave the WB at whatever.

If you're shooting for a quick (and dirty) videoworkflow, this is not a good option...

my 2,5 c...
 
Jason Ing said:
However, I recently spoke with Michael Cioni of Plastercity at NAB and he has the opposite opinion. Because I respect his knowledge, it did cast some doubt and I've resigned myself to waiting for my own testing.

Edit: But then again, I highly respect Graeme's knowledge, so I'm content with being in the ETTR camp.
Click to expand...

Guys, there is a simple way of figuring out why "expose to the right works" when you have the liberty of moving a smaller dynamic range across your latitude space. What ever written below is a simple illustration and not meant to be taken as an exact process.

Suppose your small dynamic range signal is S units and your place it somewhere along your full range of latitude. Assume the camera temporal noise is N. So you have:

signal + noise = S + N. ---------------------- (I)

Now, assuming linearity you exposed the sensor more by 10 times without clipping (i.e., exposed to the right) and now your signal becomes 10 * S. So,

Signal + noise = 10 * S + N.

Now, you can recover the original signal by dividing by 10 times.

(10 * S + N) / 10 = S + N / 10. --------------- (II)

If you compare (I) and (II), you see that you have kept the same signal but reduced the noise by 10 times!
 
This is fine, but it seems like a lot get mixed up and think that this is also true when they work from the proxies, or batch processed prores files.

If you follow that lane, you're basically working with a very nice videocam with 35 mm dof. Not a RAW cam. And definitely nothing you need to take into some uncompressed stage...


That was sorta my point:
Know the workflow you're shooting for...
there's no absolute right or wrong, just mixups...
 
Irri said:
Say you shoot something very flat, for example only using 1/3 of the sensor's dynamic range,
Click to expand...

Hi,

If you do that the final result is likely to look very boring, however it may be technically correct.

Find the sensors full dynamic range by testing, then you will know where to place the exposure! BTW the color temp of the light has an effect too.

YMMV

Stephen
 
Stephen Williams said:
Hi,
Find the sensors full dynamic range by testing, then you will know where to place the exposure! BTW the color temp of the light has an effect too.

Stephen
Click to expand...

YUP on the actual clipping of the sensor, that's why we're asking for exposure meters inserted before the metadata are applied...

If you're posting without a "development" stage, the color temp/wb settings in camera is indeed affecting your shot a lot. If you're shooting for RAW, it's only metadata.
 
gunleik said:
Y If you're shooting for RAW, it's only metadata.
Click to expand...

Hi Gunleik,

My understanding that the 'traffic lights' don't actually indicate clipping on the sensor, they are influenced by what color temp is indicated on the camera. Therefore I think it's misleading to say 'it's only metadada'.

Stephen
 
We do agree... :)

There is a curve + a color temperature setting applied to the preview, and that is what is measured, which means that the traffic light is not accurate (they are very nice, though!), and less so when you're not shooting under the sensors native ballance (which is ca 5600k)

But the sensor itself doesn't change due to these controllers being applied to the preview... So it would be good to have the exposure metering (i.e. traffic lights and waveforms) applied before metadata is applied. With the metadatasettings, the exposure meters can actually be a bit off...

Say (theoretically) you set your cam to shoot @ 500 ASA @ 5600k
The traffic lights would warn you, and you'd stop down, but in fact you have the headroom between 320 and 500, so you'd be tempted to underexpose.

The reverse is also true.

And this is the same thing that happens with the metering if you go far off from 5600k. The channels clip individually, and the metering will be off.

Sorta according to how I understand it/what I've seen in post.

But this is all assuming you shoot for a RAW and not a video workflow...

If you're planning to skip the development stage, everything is quite accurate from what you see and read on the meters.

Ideally the metering would be selectable: Shoot for RAW/Shoot for immediate processing

Makes sense?
 
Exposure notes

Exposure notes

Unless you are absolutely, positively going to do the RAW corrections and post production personally I highly recommend you set your white balance as accurately as possible in the field. The full understanding of the concept "its only metadata" is very likely to be lost on at least one critical person who deals with the footage after it leaves your hands. Play the percentages and do your part to quell the RedOne FUD.

Now about the exposure sweet spot: the axiom of exposing for the highlights (ie, avoiding clipping) and lighting for the shadows is my personal suggestion. Joofa makes a good point about ETTR improving S/N ratios. The RedOne can make spectacular images but its greatest challenge remains noise in the toe, particularly if your lighting environment does not feed the blue channel.

When working with DPs unfamiliar with the RedOne I always suggest plenty of footcandles, 5600K sources and keep the ratios a little flat during acquisition. It is very simple to tweak the curve in RA/RC/Scratch/etc to hit the intended key/fill ratio and you quell quite a bit of noise while you're at it. This does truncate absolute DR slightly but I find the tradeoff worthwhile. YMMV.
 
traffic light and exposure

traffic light and exposure

As I understand it, the traffic light is based on the Rec 709 conversion and not what's happenning on the sensor, isn't that correct?

The whole ETTR thing is something I've been in discussions/agurements with DP's and it's like suggesting something obscene to them. It always comes down, from their persprective, to "you need to protect the hightlights, and give it that nice film like curve, as digital sensors clip highlights hard."

Sure I would never advocate destroying the highlights.

But to my mind, That basically means underexposing and raising the levels in post, which creates some nasty noise in the blacks. And if you forced to me to choose from one devil over another, I'd take the clipped highlights vs the noisy blacks.
 
lillbullen said:
In the video world the brightness and light is messured in watts. And the visible part of the image is messured from 0,3 W to 1,0 W. But when your making an image brighter in post your pulling brightness up from under the 0.3 W. So you still got information from the brightness range that you can't see.
Click to expand...

Not quite correct, brightness or luminance is measured in volts not watts, from .3V (black level) to 1V (peak white). As with anything over 1V being clipped, anything under .3V is "crushed" and can not be brought back up. Also with regards to ETTR, that works for DSLR's and RED but you'll be in trouble if you use that theory for a standard video camera. If you've ever noticed what happens when you push skin tones much above 60 or 70% depending on how your gamma is set, the skin will start looking quite unnatural well before it clips at 100%.
 
You must log in or register to reply here.
Back
Top