Sony tells DSLR shooters they're idiots

Chris Malcolm wrote:
In rec.photo.digital Wolfgang Weisselberg wrote:
Chris Malcolm wrote:
In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:

Which is why autoexposure has kept improving as the number of
sensors has increased.

Do you have any proof for that? Last I heard the wins were
negible to marginal, but I could have heard wrong ...

It was a generalisation based on the fact that in my sequence of
camera purchases autoexposure has improved as the number of sensors
has increased, and seeing others report similar experiences.

In the same time the world population has also increased, as
have the national debt of the US. Based on that fact higher
national debt and more people improve autoexposure. :-)

Except that we have no reason to expect that world population and
national debt affects autofocus,

Oh, the higher the population density, the higher the chance
that any random focus will hit some person ...

whereas the number of autofocus
sensors was increased in order to improve autofocus. That doesn't
prove it, but it makes the causal relation more likely.

.... but what does that have to do with auto*exposure*?


Of course once there are enough sensors to be able to find faces and
figures within the image there are further opportunities for more
subtle categorisations, e.g. the photographer probably wants the
detail in frontal faces to be visible. The user can also control the
camera's rules, e.g. if face recognition is turned in the menu, the
camera will expose for the faces it finds, otherwise not.

Really? Any proof? At least for now, you can only switch
face recognition on/off for AF --- which need not have any
influence on AE.

It *need* not, but in cases where the kind of autoexposure is biassed
towards the in focus areas it inevitably will be as a natural
consequence of the autofocus following the faces.

I'm afraid that while your claim is plausible, it doesn't mean
it's true. One *easy* counter example: AF face recognition on &
integral exposure metering set.

That's not a counter example to my claim. It's a counter example to
general claim I quite sepcifically did not make. Read my claim again.

"e.g. if face recognition is turned in the menu, the
camera will expose for the faces it finds, otherwise not."
^^^^

Sounds like a definite claim to me.

Note too BTW that in some recent DSLRs you can no longer set purely
integral metering. The handbook may suggest you can, but in practice
there's a bias towards any focus sensors which are in use.

What part of that means face recognition plays any role in
metering in that case?

IIRC this
has been discussed here in recent monts -- unexpectedky large shifts
in exposure when slight changes in composition move a bright area off
the sensor focus area.

That would indicate a strong bias. Or a user error.

Same goes for the RAW files of some cameras
which turn out to have been unavoidably slightly cooked.

You mean like Nikon's long exposure median filter? (Google
for "Nikon mode 3") That's not even nearly recent.

-Wolfgang

In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:
In rec.photo.digital Wolfgang Weisselberg wrote:
Chris Malcolm wrote:
In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:

Which is why autoexposure has kept improving as the number of
sensors has increased.

Do you have any proof for that? Last I heard the wins were
negible to marginal, but I could have heard wrong ...

It was a generalisation based on the fact that in my sequence of
camera purchases autoexposure has improved as the number of sensors
has increased, and seeing others report similar experiences.

In the same time the world population has also increased, as
have the national debt of the US. Based on that fact higher
national debt and more people improve autoexposure. :-)

Except that we have no reason to expect that world population and
national debt affects autofocus,

Oh, the higher the population density, the higher the chance
that any random focus will hit some person ...

whereas the number of autofocus
sensors was increased in order to improve autofocus. That doesn't
prove it, but it makes the causal relation more likely.

... but what does that have to do with auto*exposure*?

Sorry, brain fart, kept saying autofocus when I meant autoexposure.

Of course once there are enough sensors to be able to find faces and
figures within the image there are further opportunities for more
subtle categorisations, e.g. the photographer probably wants the
detail in frontal faces to be visible. The user can also control the
camera's rules, e.g. if face recognition is turned in the menu, the
camera will expose for the faces it finds, otherwise not.

Really? Any proof? At least for now, you can only switch
face recognition on/off for AF --- which need not have any
influence on AE.

It *need* not, but in cases where the kind of autoexposure is biassed
towards the in focus areas it inevitably will be as a natural
consequence of the autofocus following the faces.

I'm afraid that while your claim is plausible, it doesn't mean
it's true. One *easy* counter example: AF face recognition on &
integral exposure metering set.

That's not a counter example to my claim. It's a counter example to
general claim I quite sepcifically did not make. Read my claim again.

"e.g. if face recognition is turned in the menu, the
camera will expose for the faces it finds, otherwise not."
^^^^

Sounds like a definite claim to me.

It is a definite claim. I didn't say it wasn't. I said it wasn't a
general claim. Which it isn't. The point is that a general claim can
be disproved by a single counter example. Your counter example
countered a general claim I did not make, not the specific and
definite claim I did.

Note too BTW that in some recent DSLRs you can no longer set purely
integral metering. The handbook may suggest you can, but in practice
there's a bias towards any focus sensors which are in use.

What part of that means face recognition plays any role in
metering in that case?

Because, as was mentioned in earlier posts, face recognition being
switched on will cause preferential selection of the focus sensors
which best cover the face. So if the camera has an inherent and
ineradicable bias in exposure towards the selected focus areas (as
some do) this will become a face-biassed exposure.

IIRC this
has been discussed here in recent monts -- unexpectedky large shifts
in exposure when slight changes in composition move a bright area off
the sensor focus area.

That would indicate a strong bias. Or a user error.

The posters reporting these effects have claimed an unexpectedly
strong bias. Other have suggested user error. That's always possible,
but having seen it myself and carefully tested it to verify that it's
the camera doing it even when unbiassed full frame exposure metering
has been selected I don't doubt that some cameras do behave like that.

--
Chris Malcolm

Chris Malcolm wrote:
In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:
In rec.photo.digital Wolfgang Weisselberg wrote:
Chris Malcolm wrote:
In rec.photo.digital.slr-systems Wolfgang Weisselberg wrote:
Chris Malcolm wrote:


That's not a counter example to my claim. It's a counter example to
general claim I quite sepcifically did not make. Read my claim again.

"e.g. if face recognition is turned in the menu, the
camera will expose for the faces it finds, otherwise not."
^^^^

Sounds like a definite claim to me.

It is a definite claim. I didn't say it wasn't. I said it wasn't a
general claim. Which it isn't. The point is that a general claim can
be disproved by a single counter example.

*Any* claim can be disproven by a single counter example that
fits.

Your counter example
countered a general claim I did not make, not the specific and
definite claim I did.

Assume face recognition (for AF, since there is no AE face
recognition switch) is turned on in the menu. Your claim:
THEN the camera WILL expose for the faces it finds.

My counter example: ... even when set to integral (or centre
spot, assuming the face is off center, for that matter)?

Easy to test:
- Centre spot: put a bright lamp in the center spot and a well
underexposed face to the side, set AE to centre spot, let the
AF capture the face, see what you get.
- Integral: darkish room, brightly lit face in small part of the
frame and to the side. Set AE to centre weighted integral,
let AF capture the face, see what you get.

If the face is well exposed and the room/lamp not, you've got
a broken camera, IMHO, otherwise your camera doesn't expose
for the face even though your claim says it does.

Not having your camera I find it a bit hard to test your
camera's behaviour in that point.

Note too BTW that in some recent DSLRs you can no longer set purely
integral metering. The handbook may suggest you can, but in practice
there's a bias towards any focus sensors which are in use.

What part of that means face recognition plays any role in
metering in that case?

Because, as was mentioned in earlier posts, face recognition being
switched on will cause preferential selection of the focus sensors
which best cover the face. So if the camera has an inherent and
ineradicable bias in exposure towards the selected focus areas (as
some do) this will become a face-biassed exposure.

That's assuming a lot ('if the camera has') and is quite some
backpaddeling from your former 'will'. Agreed, IF the camera
does not honor spot or centre AE modes and IF the camera does
weight the focus point highly and IF you cannot disable that,
then the camera will expose at least somewhat for the face.


IIRC this
has been discussed here in recent monts -- unexpectedky large shifts
in exposure when slight changes in composition move a bright area off
the sensor focus area.

That would indicate a strong bias. Or a user error.

The posters reporting these effects have claimed an unexpectedly
strong bias. Other have suggested user error. That's always possible,
but having seen it myself and carefully tested it to verify that it's
the camera doing it even when unbiassed full frame exposure metering
has been selected I don't doubt that some cameras do behave like that.

That's probably the reason some people do call matrix metering
unreliable and "rolling dice". But they did that even with
cameras that lack the sensors and algorithms to face detect
and some cameras also show that behaviour when in fact no face
is within the frame.

Thus: no proof that face detection plays a role.

-Wolfgang