April 21, 2014, 10:05:35 AM

Author Topic: Exposing to the left?  (Read 5583 times)

sanj

  • 1D X
  • *******
  • Posts: 1272
    • View Profile
Exposing to the left?
« on: February 24, 2013, 11:09:25 AM »
Hello all.
I find that both my 1dx and 5d3 expose a bit to the left and not to the right (recommended by many).
Am I mistaken or any of you also noticing this? 1dx does this bit more than 5d3.
I work in aperture priority.
Should I dial in slight over exposure in the compensation menu or leave it as is and boost in PS?
Thank you.
Sanjay
PS-I am spending next ten days in Masai Mara and will go to lengths to find internet to see your helpful comments. So please respond at your earliest! :)

canon rumors FORUM

Exposing to the left?
« on: February 24, 2013, 11:09:25 AM »

bdunbar79

  • 1D X
  • *******
  • Posts: 2430
    • View Profile
Re: Exposing to the left?
« Reply #1 on: February 24, 2013, 11:44:51 AM »
Yes, 1DX does and more so than the 5D3 and I noticed this on both bodies.  For instance, everything equal, at ISO 2000 the 5D3 image will be brighter than the 1DX image.  It's well-known among most of the 1DX users I know.  You can do one of two things.  You can either shoot more to the right, or "fix" it relative to cameras you are used to using in the past.  Page 325 of the Manual, under AE Microadjustment (top explanation) will tell you how to do this.  AE Microadjustment is in 1/8 increments, and automatic metering is currently being done to 0 EV.  If you'd like it to meter relative to 1/8 EV, go ahead and up it to 1/8.  I set one of mine to 5/8 EV as a test and didn't have to ETTR much anymore.  But I'm not used to that vs. past cameras, so I put mine back at 0, and chose option 1 as simply getting used to the metering and now I just shoot at +1 to +1 2/3 EV if I can. 

sanj

  • 1D X
  • *******
  • Posts: 1272
    • View Profile
Re: Exposing to the left?
« Reply #2 on: February 24, 2013, 12:01:26 PM »
Thank you so much. I will do the same.
Although am contemplating if it is better to not risk clipping highlights by increasing exposure compensation. I could open up the blacks in PS a bit if I find the photo bit dark.
But I think I will end up increasing exposure value by 1 point.
Appreciate!

bdunbar79

  • 1D X
  • *******
  • Posts: 2430
    • View Profile
Re: Exposing to the left?
« Reply #3 on: February 24, 2013, 12:26:02 PM »
+1EV should work fairly well.  I use CWA for meter readings and typically +1 works well for a subject isolated from background.  For total scene brightness, I think +1 would be a good starting point for any metering method.  Good thing about the 1DX is that files can be pushed heavily, so clipped highlights with that camera in post are absolutely no problem, although annoying that you have to deal with them.  So if you did accidentally do that, much easier to deal with using a 1DX vs. most other cameras.

TrumpetPower!

  • 1D Mark IV
  • ******
  • Posts: 951
    • View Profile
Re: Exposing to the left?
« Reply #4 on: February 24, 2013, 12:43:15 PM »
Digital exposure is a complex topic, surrounded by lots of confusion and clouded by obfuscation.

The sensor records light in a linear fashion. For each pixel, for each exposure, a number is reported from 0 to 16,384 (typically; there are other possibilities). Each pixel has either a red, green, or blue filter in front of it and so each pixel records (mostly) only red, green, or blue photons. (The filters relatively wide, mimicking the sensitivity of the nerves in our eyes. A very few blue photons will make it through the green filter and even fewer through the red filter, and vice-versa. And blue-green photons will make it through both the blue and green filters, but not as many as blue will make it through blue and green through green.)

Exactly how many photons it takes to result in a pixel recording, say, the number 1126, will vary from sensor to sensor, but exactly twice as photons will cause the sensor to record 2252.

Further, prior to that number getting recorded, different amounts of electronic amplification gets applied to the signal before recording. The number of photons that causes the camera to record 2252 at ISO 100 will cause it to record 4505 at ISO 200, and 9008 at ISO 400.

Just as turning up the volume knob on your stereo causes more and more distortion, so does the electronic amplification of the camera's ISO setting. What in a perfect world would result in recording 9008 at ISO 400 might result in 8996 in one pixel and 9012 in a pixel next to it, both of which received the exact same number of photons.

Now, let's assume you're in a perfectly-controlled studio environment with a flat subject and even illumination. And, you're photographing a scene that includes a light trap (a hollow black-lined box with a small hole at the top), an 18% gray card, and a piece of Teflon thread tape (which reflects 99.9% of the light that hits it).

Your picture, if perfectly exposed, would record 0 for the light trap, 2^14 * 0.18 = 2949 for the gray card, and 16384 for the tape.

But.

Remember those filters? They really mess things up.

First, "white" light (which doesn't exist, but never mind) is a mixture of light of all frequencies, but not in equal proportions. Second, for very good technical reasons, your camera has as many green-sensitive pixels as it does red and blue pixels combined.

So, the perfect exposure now becomes one in which the brightest channel (which is virtually always the green channel) has those values indicated above and the other two channels fall where they do. Typically in daylight, the blue channel will be about 2/3 stop underexposed relative to green, and red will be just over a stop underexposed relative to green. Those ratios shift depending on the light source, and figuring out exactly what those ratios are is what white balance is all about.

So, let's say that we nailed the in-camera exposure and the gray card came out at 2949 in the green channel. The red channel might have come out at 1475, and the blue at 1946. Presumably, the light trap still came out at 0 for all three channels, and the thread tape would have come in at R=8192 G=16384 B=10912.

To get your perfectly white balanced perfect exposure, you'd double all the red values, leave the green values alone, and multiply all the blue values by 3/2 -- after which the red, green, and blue values for the three objects under discussion (all of which are different shades of gray) would be the same.

But.

What about the ball bearing in the scene? Much of it is just mirroring the rest of the scene, but there's that specular highlight that's reflecting the light source, and that part is a lot brighter than the thread tape. In our photo, all the parts of the ball bearing that are exactly as bright as the thread tape get recorded as R=8192 G=16384 B=10912, but the parts brighter...well, green can't get any brighter, so it stays at 16384, but let's say that R=10912 and B=14549. The camera is telling us that it's a lot more magenta than it really is. And an even brighter spot (but not the brightest spot) has all three channels maxed out at 16384. We're back to white, but it's telling us that this is the same white as the thread tape, which it clearly isn't. And what about this other part of the scene where we've turned up the brightness on our lighting?

...and that brings us to how digital exposure actually works in the real world.

Specifically, all camera meters on the market actually tell you to underexpose the scene. My 5DIII, for example, underexposes by about 2/3 stop. The raw processing software (both in-camera and Lightroom, etc.) then applies an equivalent amount of digital overexposure to compensate. You could make a spreadsheet of all the values recorded by the sensor, multiply by (roughly) 3/2, then multiply red and blue by whatever you need for white balance, and the numbers would (basically) match the numbers coming out of the JPEG.

What that does is give you an extra 2/3 of a stop of headroom for your highlights. It also means that something more sophisticated than a simple linear multiplication can preserve more visible detail in those highlights.

There's no such thing as a free lunch, of course. The worst noise is always in the shadows, and applying 2/3 stop of digital boost will make those shadows that much noisier. All things considered, though, those shadows are awfully clean these days, so the tradeoff of having a bit more headroom for highlights is well worth a bit of invisible noise in shadows.

There are some additional caveats. Virtually all RAW processors (including the one in the camera that makes JPEGs) apply all sorts of other modifications to the data before you see the image. Most significant is the gamma adjustment; the sensor records in linear gamma, but the rest of your workflow is set up for gamma 2.2. Linear gamma images displayed without proper adjustment look very dark and contrasty. Next most significant is color profiling, which is yet another can of worms. And then there's an s-curve generally applied for contrast and "pop," there's almost always some sort of "secret sauce" (the picture style) to give a certain "look," and then there's all the knobs that you can fiddle with yourself (contrast, saturation, shadow boost, highlight recovery, and the rest). The end result is a very complex mathematical transform applied to the data.

And because that transform is so complex, that leads to the most unfortunate factor of all. Applying post-exposure digital exposure adjustments anywhere other than to the initial RAW recording of the data is going to amplify and otherwise interact with all those other adjustments, and generally result in chaotically unpredictable behavior. Modest changes generally don't have much of a visible effect, but the results with more dramatic changes can be...well, more dramatic.

I don't know if Camera Raw / Lightroom apply their exposure adjustments (with that top slider) before or after all the rest of the calculations. I would hope they're smart enough to do it before, but I've long since given up on ACR for color-critical work. I know for certain, for obvious reasons, that Photoshop itself is going to do anything like that after everything else has been done.

So, my recommendation for a general-purpose workflow is to target your exposures to the same as what the camera's meter is set to; that's almost guaranteed to be the best compromise between preserving highlights and reducing shadow noise. That's not to say that you should blindly trust the camera's meter, of course; meters can easily be fooled. Rather, get to know how your meter works with a perfectly-lit gray card and try to achieve that level of exposure. How to do that in practice is the usual challenge of the photographer...perhaps you'll use an external meter calibrated to your camera's meter, perhaps you'll use Adams-style spot metering, perhaps you'll gauge the histogram, perhaps you'll judge it from the preview image, whatever. But the point is to learn how to get the camera to expose the scene the same way its meter would under ideal metering circumstances.

If you can do that, you'll get the best compromise of preserving highlights and reducing shadow noise that your camera is likely capable of, and you won't have to deal as much with all the mathematical funkiness that goes on behind the scenes with digital development. Your images will be basically right straight out of the camera.

In scenes where you're still blowing highlights that you care about and have excessive noise in the shadows, you should first strive to fix the light by any and all traditional photographic means -- wait for the Golden Hour or add fill flash or use scrims or reflectors or whatever. If that won't work, either HDR of some form or a larger format camera is your ticket.

With a particularly convoluted workflow in controlled environments, you can more intelligently apply the "ETTR" concepts...but, even then, you're generally best off adjusting exposure as I described above so that the green channel is right on the 1.0 gamma graph. How to do that is much more involved than is reasonable for this forum...it involves analyzing ICC profiles built from a linear UNIWB development of the RAW image, and software (such as Raw Photo Processor) that lets you specify the channel multipliers yourself....

Cheers,

b&

P.S. If you were to shoot, for example, a ColorChecker exposed exactly according to your camera's meter reading of a gray card, and it doesn't look almost identical to an idealized ColorChecker such as you can find at Bruce Lindbloom's site, then you've got a problem somewhere with your camera's meter. But if both your photo of the ColorChecker and Bruce's look too dark, then it's much more likely that your monitor's brightness is off, or that something other than the camera is to blame. b&

Viggo

  • 1D X
  • *******
  • Posts: 1696
    • View Profile
Re: Exposing to the left?
« Reply #5 on: February 24, 2013, 12:58:23 PM »
Yeah, the metering was a big issue for me with the 5d3, I adjusted and adjusted so almost every shot had to be shot twice.

The 1d x was better, at least much more consistent no matter what lighting situation I have, but mine is dialed in, through custom function , to offset 0 ev to +6/8 i believe. It works great and i have never had a shot where i lost highlights and thought it was a bad idea to do the offset.
« Last Edit: February 24, 2013, 01:00:32 PM by Viggo »
1dx, 17-40 L, 24-70 L II, 85 L II, 200 f2.0 L

Plamen

  • Canon AE-1
  • ***
  • Posts: 78
    • View Profile
    • Math and Photography
Re: Exposing to the left?
« Reply #6 on: February 24, 2013, 01:02:27 PM »
My 5D2 underexposures slightly as well. With a flash and shiny objects, it underexposes even more. I believe that the metering system is too protective of the highlights. Just dial in EV= +1/3 or so.

In my experience, Canon crop bodies meter differently.

canon rumors FORUM

Re: Exposing to the left?
« Reply #6 on: February 24, 2013, 01:02:27 PM »

bdunbar79

  • 1D X
  • *******
  • Posts: 2430
    • View Profile
Re: Exposing to the left?
« Reply #7 on: February 24, 2013, 01:40:55 PM »
Yeah, the metering was a big issue for me with the 5d3, I adjusted and adjusted so almost every shot had to be shot twice.

The 1d x was better, at least much more consistent no matter what lighting situation I have, but mine is dialed in, through custom function , to offset 0 ev to +6/8 i believe. It works great and i have never had a shot where i lost highlights and thought it was a bad idea to do the offset.

Yes, exactly.

bdunbar79

  • 1D X
  • *******
  • Posts: 2430
    • View Profile
Re: Exposing to the left?
« Reply #8 on: February 24, 2013, 01:43:53 PM »
I think we're getting too complicated here for a simple problem.  The fact is that the 1DX and 5D3 meter differently.  Everybody who owns both a 5D3 and 1DX knows this.  At equivalent settings and in the same exact situation, the 5D3 files are brighter than the 1DX files.  Plain and simple.  It doesn't matter how or why, the important thing is that you know this when you take your photos and adjust appropriately if you are not satisfied with this.  That's all I did and have since moved on...way on.

TrumpetPower!

  • 1D Mark IV
  • ******
  • Posts: 951
    • View Profile
Re: Exposing to the left?
« Reply #9 on: February 24, 2013, 02:13:27 PM »
At equivalent settings and in the same exact situation, the 5D3 files are brighter than the 1DX files.

It'd be nice to know if the raw data in the files differ, or if the differences happen elsewhere.

Anybody with both cameras and suitable software (dcraw, Rawnalize, Raw Photo Processor, etc.) and a couple spare moments care to chime in?

And, just to be sure, this means ignoring what the meter says and setting the exact same lens / aperture / shutter / ISO on both cameras with the lighting held constant....

Cheers,

b&

bdunbar79

  • 1D X
  • *******
  • Posts: 2430
    • View Profile
Re: Exposing to the left?
« Reply #10 on: February 24, 2013, 02:24:20 PM »
At equivalent settings and in the same exact situation, the 5D3 files are brighter than the 1DX files.

It'd be nice to know if the raw data in the files differ, or if the differences happen elsewhere.

Anybody with both cameras and suitable software (dcraw, Rawnalize, Raw Photo Processor, etc.) and a couple spare moments care to chime in?

And, just to be sure, this means ignoring what the meter says and setting the exact same lens / aperture / shutter / ISO on both cameras with the lighting held constant....

Cheers,

b&

Yes, very interesting thing.  I was hoping you would know exactly why this could be, but it's very strange.  Also though, this is the clincher:  Push to ISO 10,000 on both cameras, and now the reverse is true.  The 1DX file is brighter.  I'm interested again in this and I'd like to do some test shots before I draw conclusions.  I'm going to do them tonight and unfortunately I'll have to convert to jpg, but I'll post them and won't process them.  Thanks.

TrumpetPower!

  • 1D Mark IV
  • ******
  • Posts: 951
    • View Profile
Re: Exposing to the left?
« Reply #11 on: February 24, 2013, 02:58:01 PM »
I'll have to convert to jpg, but I'll post them and won't process them.


That probably won't tell you as much as you might think it would.

If you have anything that can dump the raw file with UNIWB and linear gamma, that would tell us something. The classic tool for that sort of thing is dcraw, which is free but it's a command line tool.

If you're on a Mac, you could use Raw Photo Processor:

    http://www.raw-photo-processor.com/RPP/Overview.html

There's a preset for UNIWB in the white balance drop-down, a text box to enter a gamma (default is 2.2; set it to 1.0), colorimetric gamma for the curve type (default is film-like), and save it as a raw untagged TIFF (at the bottom of the list next to the Save button).

The output will be very dark and very green, but it's what the sensor is actually recording.

You could then eyeball the two side-by-side, or you could scale them to the same pixel dimensions, layer the one on top of the other, and set the blend mode to "difference" to get a visual map of how much they differ by.

If the latter results in basically a solid black image (save for registration errors), then the cameras are behaving close enough as makes no difference and the differences you're seeing come from something in the raw processing chain.

If, however, there actually is a difference, then the cameras actually are exposing differently.

If you like, if you can make the raw files available to me, I'll have a look at them myself....

Cheers,

b&

TrumpetPower!

  • 1D Mark IV
  • ******
  • Posts: 951
    • View Profile
Re: Exposing to the left?
« Reply #12 on: February 24, 2013, 03:13:39 PM »
I'm going to do them tonight and unfortunately I'll have to convert to jpg, but I'll post them and won't process them.  Thanks.

Incidentally, if you can include a ColorChecker Passport in the shots, I can give you some hard empirical figures....

b&

canon rumors FORUM

Re: Exposing to the left?
« Reply #12 on: February 24, 2013, 03:13:39 PM »

Mt Spokane Photography

  • 1D X
  • *******
  • Posts: 7714
    • View Profile
Re: Exposing to the left?
« Reply #13 on: February 24, 2013, 03:27:26 PM »
Metering is not simple because it depends on the subject.  A snow scene will be badly undersxposed, but a scene with all dark tones may be overexposed.  Its all within a reasonable tolerance of about 1-1/2 stops, but there is seldom a perfect exposure because real scenes conntain different light and dark objects and you are working with averages.

TrumpetPower!

  • 1D Mark IV
  • ******
  • Posts: 951
    • View Profile
Re: Exposing to the left?
« Reply #14 on: February 24, 2013, 03:42:56 PM »
Camera manufacturers peg where the meter actually meters to, to do this they have their own calculations for iso value.

Once they have done the iso calculation the meter still doesn't meter to a midtone, it is NOT 18% grey reflectance as we expect, the ANSI standard ( American National Standard for General-Purpose Photographic Exposure Meters (Photoelectric Type)" ANSI PH3.49-1971 ) calculation differentiates between reflectance and incidence light. When all is said and done the meters actually meter to around 12-13% grey reflectance, or 2/3rds of a stop underexposed.

http://www.richardhess.com/photo/18no.htm
http://www.bythom.com/graycards.htm


Actually, I don't think we're discussing metering but actual exposure.

That is, ignore the camera's onboard meter entirely. Step outside on a sunny day and take a shot at 1/100s @ f/16 @ ISO 100 with both cameras. (Or, preferably, in a studio with a more controlled environment.) What does the actual unprocessed raw data look like for each? Let's say you've got a ColorChecker in the scene. What are the raw RGB values of the N5 patch for each before any processing (including before white balancing)?

Cheers,

b&

canon rumors FORUM

Re: Exposing to the left?
« Reply #14 on: February 24, 2013, 03:42:56 PM »