Metering mode for portraits

It’s a tool and I’m OK having learnt to use it (I don’t know how the algorithms in my computer work either) by experience.

A large part of this came from my years with a 4x5 film camera. I was religious for several years Sekonic 2° spot metering the high and low ones in my scenes and figuring the appropriate exposure - transparencies don’t give you much freedom! I also used a crop camera (10 and 20D for most of it) with a 17-40 that nicely spanned the range of FL of my 4x5 lenses to scope out compositions, camera positions and angles etc - transferred the FL to the right lens. It was extremely rare that the exposure suggested by evaluative from the digital camera was more than 1/2 stop off what I had measured with the Sekonic.

These days, I tend to shoot to the right if I can and am either watching the live histogram and/or checking for blinkies while overriding the evaluatiive - not because I don’t trust it but to get to the right to be able to use all the DR of the sensor: unless you ETTR, you are usually not getting all the DR available. Each uses the tools to their own advantage - the key is knowing each tool and how to use it well.

Yet mere adherence to Evaluative programming yields photos like this one, in which the AF point was on the 18% gray card, and although the Evaluative mode theoretically biased its exposure to that AF point (while also factoring in the other zones) this is what resulted...underexpose​d gray card in the photo. I wish I could predict how much to bias that shot with EC, to make the gray card and the white railing both at their inherent brightnesses (and not so dark as in the photo), but I have no idea how much the adjacent metering zones wrongly bias the 'priority AF zone'.

Yup - unthinking use of a tool can lead to poor results. Did you want a nice photo of the sky or of something in the foreground? Evaluative has probably been trained to think the former if there is a large amount of sky in the frame. I sometimes find myself pointing the camera down a bit to reduce the amount of sky in the evaluation in such instances and use that info to correct with - a bit like focus and recompose, only with exposure.

The top image looks like what I would see before taking off my sunglasses (and may be the image I want) - and the bottom one is the “after” shot: too bright for me The evaluative decided to give the sky more weight because of how much of the frame it occupied. Both choice of exposure and how to get it are up to the operator. I might also end up with the bottom one, however, if the sky’s not blown, because it would be ETTR’d - the final image TBD in PP Horses for courses, I suppose

davesrose wrote in post #18547541
I still don't find that wording true either: they didin't "reduce" photos into the programming. They "interpreted" photos to program all the scenarios the meter would measure. The array's scenarios would be values of the direct measurements from the meter.

But that is the same thing. For an evaluative meter to work, when using rule-based processing of the incoming data, you have to locate a scenario with a similar pattern in your database.
An example: Say we have a two zone meter. You get metering results that are EV 11 and EV 13. Reduced to a common baseline that's (0,+2)|+11, which should be interpreted as the right zone is two stops above the left, and the left is at EV 11.

Now you locate a pattern for the difference (0,+2). The data entry associated with that pattern is +1.3, which in this case means expose with +1 1/3 of a stop, relative the left zone, and you are good. Your end result is a setting at EV 12.3. A lot of images where the right half is two stops above the left will be condensed into the same (0,+2)+1.3 database record.

Since you have more zones in a real situation, you need a clever search algorithm to find the associated database record quickly enough. Perhaps a hash algorithm will do.
But exactly how it's implemented is not important. You can take my example as a possible way of reducing a lot of data to a manageable size.

AJSJones wrote in post #18547816
I was religious for several years Sekonic 2° spot metering the high and low ones in my scenes and figuring the appropriate exposure.

I'm far from an expert and I'm here to learn, but if it were me wanting to ETTR (which I do) I'd be using that Sekonic 2° spot meter (which I don't have) to meter the lightest bit I don't want blown and adding 3 stops. Then I'd adjust to taste with a raw converter. That's what I do if I have time but my camera's spot meter is only 5° so it's harder to get even light across it.

Wilt wrote in post #18547818
Had I been shooting a photo of someone, their face would be dark as exposed by Evaluative.

Which to the experienced photographer that uses evaluative, and does not have a blind faith in evaluative, they also know the situations where partial or spot metering are more appropriate (back-lighting being a prime example). As for why evaluative didn't take more weight in the AF point for your "underexposed" photo, my guess is that evaluative still takes the whole scene's exposure into account when exposure range is not being exceeded.

apersson850 wrote in post #18547835
But exactly how it's implemented is not important. You can take my example as a possible way of reducing a lot of data to a manageable size.

Yes, but my point was that data is based on scenarios from the values produced from the meter. Photographs are analyzed before programming (and patterns can be deduced), but they are not directly digitized or compacted into the meter's software.

Pippan wrote in post #18547838
I'm far from an expert and I'm here to learn, but if it were me wanting to ETTR (which I do) I'd be using that Sekonic 2° spot meter (which I don't have) to meter the lightest bit I don't want blown and adding 3 stops. Then I'd adjust to taste with a raw converter. That's what I do if I have time but my camera's spot meter is only 5° so it's harder to get even light across it.

There are indeed many ways to go about this, as I said

By the time you’d’ve gotten the Sekonic out and pointed it and read off the reading and adjusted your exposure, I’ve taken a shot with a well-educated guess (based on my familiarity with the metering mode and assessment of the scene) and have examined the histogram and blinkies to see if any further adjustment is needed. I read the camera’s histogram as a display of the thousands of brightness levels in the scene - each pixel being equivalent to a very narrow spot reading. You ony got one reading I am not saying you should do it my way (nor implying that I think you are telling me to do it your way) - just saying what works very well for me.

davesrose wrote in post #18547956
Photographs are analyzed before programming (and patterns can be deduced), but they are not directly digitized or compacted into the meter's software.

Not the photos themselves, of course. They have no meaning in a metering concept. But their light levels can be condensed and stored, so the relevant (for light metering) information in the photos are then stored in the camera. You can't recreate images from them, except for the crude representation of the image that the light levels actually are. Well, with a 252 zone metering system, maybe you can get some idea about the original image?

I suppose modern meters are essentially analyzing a small live photo (Canon states the 5D4 has 150,000 pixels....within those, segments). Canon is indicating the current software does more then just compare scenarios: but has actual "intelligent" recognition (such as facial recognition that can trace outlines of the face, and consider contrast values within that area). If Nikon claimed a collection of 100,000 photos from which to derive the FA meter's algorithms, who knows how many photos and separate studies have been used to as a basis for creating software now. Before creating the final software, it is conceivable that software engineers are sorting through photo examples to derive common patterns. So maybe semantics to say "reduce" when "interpret" is a more accurate word for the software development that occurs. But then the advertising department is a separate section entirely (and words can change like a game of telephone)

I see it as little more than computer-assisted pattern recognition, where a finite number of patterns present in the thousands of photos in the "training" database were identified and used for comparing with the "live" input from the available metering segments.