My camera is almost always in Manual with Spot metering. Knowing where to place the the metered area (on the +0- scale) based on how you want it exposed and the tonality of the subject is the most accurate method for me.
wilt wrote:
The 18% gray card is only properly exposed with Spot metering.
However, the evaluative exposure is the best exposure of the three for a backlighted subject. Evaluative determined the type of lighting you had set up and exposed properly for it.
Of course a computer can't know that you wanted a "wrong" result...Captain Kirk taught us that back in TOS.
Wilt,
You make a good point in your spot metering example, but I think that what you're really showing is why it's often bad to rely on a camera's autoexposure system, regardless of the meter pattern selected.
My strategy, most of the time, especially when shooting action, is to meter off an 'easy' target, like a patch of sunlit grass on a playing field, and then use that as the basis for manual exposure settings. That way you don't have to be concerned about the camera making bogus exposure changes for shots containing a changing mix of light and dark faces, light and dark team uniforms, etc.
aborr wrote:
Wilt,
You make a good point in your spot metering example, but I think that what you're really showing is why it's often bad to rely on a camera's autoexposure system, regardless of the meter pattern selected.
The issue is not 'autoexposure' (which merely sets one or more parameters for you) but the suitability of any metering pattern along with whether or not the object/scene being metered correlates to the meter's fundamental asssumption of 'exposure for 18% gray'. Whether you use M or Av or Tv, all three can be equally wrong in the shot, because the METERING was based wrongly upon a wrong brightness.
In the case of the Evaluative example which I posted, while the camera was supposedly biased to the focus zone (on the 18% gray card), it also factored in the brightness of the adjacent metering zones which was NOT 18% tonality!
I use Evaluative most of the time but for tricky situations prefer spot metering. The only time I use CW is when I shoot my annual roll of film retro-out with my FM3A.
It would be nice if Canon bothered documenting what "evaluative" does.
We know what Nikon's matrix does (AI techniques to match the R, G, and B reading matrix with a scene database, hence the name). There is enough in their descriptions for someone with a decent academic background in AI to get a very clear idea of how it works, and the pros and cons of using it in various situations. This can then filter out to a wider audience.
Is Canon's "evaluative" similar, but just using luminance? Or does it just choose one of a small number of sets of weights depending on criteria like "this looks like sky on top" and "this looks like it's backlit"? Does Canon seriously expect me to guess?
I use centre weighted and spot only, centre weighted in lieu of a true full averaging meter as on the original OM-2.
melcat wrote:
It would be nice if Canon bothered documenting what "evaluative" does.
We know what Nikon's matrix does (AI techniques to match the R, G, and B reading matrix with a scene database, hence the name). There is enough in their descriptions for someone with a decent academic background in AI to get a very clear idea of how it works, and the pros and cons of using it in various situations. This can then filter out to a wider audience.
Is Canon's "evaluative" similar, but just using luminance? Or does it just choose one of a small number of sets of weights depending on criteria like "this looks like sky on top" and "this looks like it's backlit"? Does Canon seriously expect me to guess? ...Show more →
I've got a copy of "Modern Photography," June 1987, in which they do a strip-down test of the Canon 650, which was the first camera with the Evaluative system. According to them, the Canon system is (or at least, was then) similar to the Nikon Matrix system.
IMO, Wilt's test confirms it. The camera deduced "backlighting" and exposed properly for it, allowing the subject to be somewhat underexposed to retain the character of the light. Evaluative is not spot metering--it should not have exposed a backlighted gray card to an 18% tone. It did exactly what it was designed to do.
RDKirk wrote:
I've got a copy of "Modern Photography," June 1987, in which they do a strip-down test of the Canon 650, which was the first camera with the Evaluative system. According to them, the Canon system is (or at least, was then) similar to the Nikon Matrix system.
Unless they disassembled the firmware, that tells you nothing. It's purely a software algorithm.
IMO, Wilt's test confirms it.
No, it doesn't. Canon might simply have programmed a test based on relative subject and background area which decided that this was a black subject on a medium background. (Think black swan against water.)
The camera deduced "backlighting" and exposed properly for it, allowing the subject to be somewhat underexposed to retain the character of the light.
This is *not* what Nikon's matrix metering does. It does not know "backlighting" or anything else photographic, but simply matches the raw data against a giant scene database (6000 scenes back in the F100 days - I nearly bought that one and had the brochure). This is a well-known AI technique, but IMO a stroke of genius from Nikon to apply it to metering.
>>I've got a copy of "Modern Photography," June 1987, in which they do a strip-down test of the Canon 650, which was the first camera with the Evaluative system. According to them, the Canon system is (or at least, was then) similar to the Nikon Matrix system.<<
Unless they disassembled the firmware, that tells you nothing. It's purely a software algorithm.
I would suggest you go to a library and check out a microfiche to read their test yourself. Back in those days, both Modern and Popular Photography had experienced camera technicians who actually disassembled cameras and lenses in a test laboratory and ran exhaustive tests on the systems. Check out what they did and what they concluded.
>>IMO, Wilt's test confirms it.<<
No, it doesn't. Canon might simply have programmed a test based on relative subject and background area which decided that this was a black subject on a medium background. (Think black swan against water.)
>>The camera deduced "backlighting" and exposed properly for it, allowing the subject to be somewhat underexposed to retain the character of the light.<<
This is *not* what Nikon's matrix metering does. It does not know "backlighting" or anything else photographic, but simply matches the raw data against a giant scene database (6000 scenes back in the F100 days - I nearly bought that one and had the brochure). This is a well-known AI technique, but IMO a stroke of genius from Nikon to apply it to metering. ...Show more →
You're right, the Nikon system is a "scene matching" system--it's not really "deducing" anything--I was speaking colloquially. However, that is what the Canon system is also supposed to be (or at least was in '87).
If you look again at Wilt's examples, the scene exposed by the evaluative system did not leave the card black--it does have tonality. Modern did much more extensive tests that illustrated how clever Canon's engineers actually were with this sysytem.
seems a poll that is lacking in options...... so another that will go un-voted upon...
metering choice seems based on what you are trying to do and what the camera is trying to do and where the compromise is for the shot....
I would also second that opinion on the non-intuitive nature of the metering icons, one of the reasons I like having the Quick control Screen....it will tell you what they are...again and again and again and does not comment on your apparent inability to learn
RDKirk wrote:
I would suggest you go to a library and check out a microfiche to read their test yourself.
Our public libraries don't stock titles like these. If you want to pursue your argument you'll have to give more details.
But I don't see how disassembling the thing can tell you much. It will tell you that Canon's sensor doesn't (or didn't) have an RGB matrix like Nikon's. Other than that, you can test it as a black box, but you don't have to do any more than use it as a camera to do that.
I should caution that I suspect Nikon reused the name "matrix metering" when they introduced the system we know as that now. I remember seeing blurb for the F5 that used the modern scheme.
However, that is what the Canon system is also supposed to be (or at least was in '87).
This type of information is what I was calling for in my original post. If it's hidden in some long-forgotten "white paper" from Canon, please point to it.
If you look again at Wilt's examples, the scene exposed by the evaluative system did not leave the card black--it does have tonality.
As you would expect from either type of algorithm.
Modern did much more extensive tests that illustrated how clever Canon's engineers actually were with this system.
The implementation is actually rather simple. The scene database is the thing.
