Metering mode for portraits

BigAl007 wrote in post #18450750
And this is the reason that mostly I have mine set for CWA to this day. With CWA I do know what the camera is going to do in pretty much any situation, since it is essentially the same as my Pentax ME Super did back in 1980 when I first had a camera with built in metering.

The problem with Evaluative is that the camera will assess a whole load of different sections of the frame, and then use some unknown algorithm to try to make a guess at what it is looking at. As the scene gets more complicated you run a higher chance of the camera taking an unexpected decision. My all time favorite metering mode though is still an external incident lightmeter, preferably with circular analogue readout, which is what I started out with back around 1974 or so.

Alan

I learned photography using virtually the same camera, a Pentax Program A, which I believe was an immediate successor of the ME Super. Center weighted average was the only option. When I moved to Canon, with an Elan 7, I tried evaluative mode, but it would often give me wonky exposures. To this day, I still use CWA most often, because I find the results to be very reliable. I do use spot metering too, but only when the light is difficult. CWA, for me, is faster to use than spot metering. I don't have to be as careful.

I'm sorry to dig up such and old discussion, but I found it interesting! Photographers have strong and varied opinions about metering!

I like partial metering for portraits of one person. The 6% meter coverage pretty much covers a face so the meter isn't fooled by different color clothing. Also works with various race people (the 5D4 has color sensitive metering so it doesn't try to make all subjects 18% gray).

cristphoto wrote in post #18545376
I like partial metering for portraits of one person. The 6% meter coverage pretty much covers a face so the meter isn't fooled by different color clothing. Also works with various race people (the 5D4 has color sensitive metering so it doesn't try to make all subjects 18% gray).

Intriguing. Could you point us to the reference which led to your underderstanding that you described as the fact that the 5DIV "works with various race people (the 5D4 has color sensitive metering so it doesn't try to make all subjects 18% gray)" ?

cristphoto wrote in post #18545376
(the 5D4 has color sensitive metering so it doesn't try to make all subjects 18% gray).

Camera meters, at least the ones that I am familiar with, are not contextually aware.

The Canon 1DX was the first to use a RGB meter and " EOS iSA (Intelligent Subject Analysis) System". It mainly works during evaluative, and is meant to get more accurate exposures if the scene is more fully green, red, etc. There's also a face detection system that will weight exposure towards a face (there isn't specific information about particular racial tones):

Canon: 1DX RGB Meter Article

The 5D4 has an upgraded 150,000 pixel RGB + IR sensor with EOS iSA System with 252 zone (18 x 14) metering

Article from 2011:
http://www.learn.usa.c​anon.com …x_rgb_meter_art​icle.shtml

...until now, every EOS camera used a light-sensitive metering sensor that essentially could only measure how much light there was. Some of the latest models, like the EOS 7D and 60D, used a multi-layer system with the ability to detect excessive red or green tones, and apply a slight correction.

EDIT: it's the same article davesrose linked. I did leave in the bit identifying which bodies got this technology

Wilt wrote in post #18450749
Without any offense intended, Evaluative is the 'non-thinking mode'...

And 'the best metering mode (in general) for portraiture is to use a handheld incident light meter to read the amount of light falling upon the SCENE and use that reading on a manually-set camera.

This last recommendation, while fine for controlled situations and appropriate for a portrait studio, is also, no offence intended, a “no thinking involved” mode - read the meter, set the camera It does, however, assume you are comfortable being restricted to the 18% grey criterion for every situation.

AJSJones wrote in post #18545814
This last recommendation, while fine for controlled situations and appropriate for a portrait studio, is also, no offence intended, a “no thinking involved” mode - read the meter, set the camera It does, however, assume you are comfortable being restricted to the 18% grey criterion for every situation.

Agreed. One must ASSume that when using incident reading, you want skin tone to appear at its inherent brightness as seen in normal light.
https://media4.s-nbcnews.com …-vid-featured-desktop.jpg


But you might NOT want that to happen, but want the face to appear as if it was in subdued light (and therefor darker than its inherent brightness)
https://video-images.vice.com …;center,center&​resize=0:*

Yet evaluative is 'no thinking mode' in that even if you TRIED to do a bit of thinking, it still would leave you guessing about how it arrived at the exposure suggestion which it gave!

Wilt wrote in post #18545955
Agreed. One must ASSume that when using incident reading, you want skin tone to appear at its inherent brightness as seen in normal light.
https://media4.s-nbcnews.com …-vid-featured-desktop.jpg

But you might NOT want that to happen, but want the face to appear as if it was in subdued light (and therefor darker than its inherent brightness)
https://video-images.vice.com …;center,center&​resize=0:*

Yet evaluative is 'no thinking mode' in that even if you TRIED to do a bit of thinking, it still would leave you guessing about how it arrived at the exposure suggestion which it gave!

Yup - so either way, the operator needs to know when and how to a) interpret the data given by the device and then 2) make adjustments depending on whether the device "knows" the operator's mind. With evaluative mode (based on algorithms derived from millions of images evaluated by many users for what they wanted to expose for - maybe not cutting edge neural networks and AI but close) and a (live) histogram with millions of reflected readings, I am quite experienced enough to handle exposures - in particular in changing light or light that is falling where I cannot get to. Plenty of thinking and adjusting going on as I do so - white birds flying in front of a dark b/g or dark features in a bright scene etc. For a studio where putting the meter next to the model's face and dialling it in to the camera is fine and reliable (as appropriate for the thread on portraits). Then you add or subtract, based on thinking and desire, for the intended image, whether it's evaluative, spot, CWA or incident . Just presenting a counter "thought" to your somewhat prescriptive-sounding advice (which incidentally, as I already agreed is, well, spot-on for portraits). (On a separate topic, if I can afford the shutterspeed hit, I tend to expose to the right while watching blinkies and histograms)

AJSJones wrote in post #18545996
With evaluative mode (based on algorithms derived from millions of images evaluated by many users for what they wanted to expose for - maybe not cutting edge neural networks and AI but close) and a (live) histogram with millions of reflected readings, I am quite experienced enough to handle exposures - in particular in changing light or light that is falling where I cannot get to.

Folks need to be careful to understand that it is NIKON that has a database of images upon which it attempted to embed some 'smarts' to the evaluative meter, but Canon's evaluative meter is not necessarily based on such a database. Canon has said very little over the decades on the subject of how its Evaluative works!

Wilt wrote in post #18545400
Intriguing. Could you point us to the reference which led to your underderstanding that you described as the fact that the 5DIV "works with various race people (the 5D4 has color sensitive metering so it doesn't try to make all subjects 18% gray)" ?

Perhaps I wasn't as eloquent as possible. What I meant to communicate is with the metering system taking color into its calculation the system would tend to be more accurate independent of subjects. Prior meters (which while accurate under most conditions) tend to make results toward 18% gray.

You think Canon does not have such a basis for their evaluative? They are also in the market to sell cameras and I doubt if their evaluative system would be allowed by HQ to fail more than Nikon's. Canon does not disclose lots of things

AJSJones wrote in post #18546009
You think Canon does not have such a basis for their evaluative? They are also in the market to sell cameras and I doubt if their evaluative system would be allowed by HQ to fail more than Nikon's. Canon does not disclose lots of things

I think there is a genuine question as to whether a 'data-mining' database driven approach (as the earlier description of the Nikon approach suggests it must be) or a rules driven 'algorithmic' approach (as is implied to be Canon's) would actually be best.

In all honesty, I have no idea whether one company is following one approach or the other company is following the other ... but they are distinct and different methodologies, each of which might work better in different circumstances.

AJSJones wrote in post #18546009
You think Canon does not have such a basis for their evaluative? They are also in the market to sell cameras and I doubt if their evaluative system would be allowed by HQ to fail more than Nikon's. Canon does not disclose lots of things

We will never know, will we? But one would think that Canon Marketing would have made specific statements along the line of 'Our algorithms are smart ones' so as potential buyers do not assume that Nikon is better than Canon in that regard. The fact that nothing is said is either

• Canon's evaluative is simply not as 'intelligent' so there is no basis for a statement to be made, or
  
• They want us to simply ASSume theirs is just as good as the competition's
  

DaviSto wrote in post #18546012
I think there is a genuine question as to whether a 'data-mining' database driven approach (as the earlier description of the Nikon approach suggests it must be) or a rules driven 'algorithmic' approach (as is implied to be Canon's) would actually be best.

In all honesty, I have no idea whether one company is following one approach or the other company is following the other ... but they are distinct and different methodologies, each of which might work better in different circumstances.

I am not convinced there is a distinction. The camera’s processor “evaluates” the image - I seriously doubt that either carries the database (from which their algorithms were derived) in the camera and that both apply criteria based on their respective databases. In other words, they basically use the same principle: Historical database of wide range of pictures -> evaluations by users for each image -> develop algorithms to match those different (styles/characteristic​s of) images - load algorithms to camera. But, that is all proprietary so we won’t know.
This result from 6 years ago provides an idea of how powerful such algorithm development had become even then - getting only one value (a good exposure) for a single image must be a piece of cake with enough input from users evaluating lots of images.