Dpreview G1 Preview Revisited

[Ehrik]

But sensor SIZE does affect image DR, I've never said anything else.
But given a sensor size, then the pixel size may not affect image DR,
particularly in as how much details you can make out from the noise N
stops below clipping in the final image, viewed at a fixed size.

snip...

Everything else being equal, if we have more, smaller, pixels, each
pixel
will have less DR. (DR defined as clipping level divided by the noise
floor.)
But if we downsize the pixels to a lower-Mp image, each new pixel will
have less noise, because the noise averages out. So the DR increases.
This is the point people are missing.

To me, there seems to be a contradiction, or I don't get it at all. :s

I re-read it and can't find any errors in what I wrote, but "everything being
equal" is of course not clear-cut, so is open for interpretation.

There are two separate issues:

1) How to compare fairly if the Mp count differs between two cameras.
Pixel DR is not fair.

2) What constitutes everything else being equal, or how does the
read noise, FWC etc. vary with pixel size.

Given a certain sensor size disregarding fill factor, let's look at a
3mp and a 12mp model for an easy pixel binning setup. While shot
noise would be higher on the 12mp sensor per/ pixel, binned (or per
area) the difference would indeed been averaged out, let's assume
completely.

Yes.

But the read noise would everything else being equal be 4
times as much, given the same quality electronics.

Let's say the 3Mp camera has a full-well capacity of 4096 electrons
and a read noise of 4 electrons. Then we have a pixel DR of 10 stops
(2^10 = 4096/4).

One notion of everything being equal would be the 12Mp pixels
having a FWC of 1024 electrons and a read noise of 4 electrons.
Then each pixel has a DR of 8 stops (2^8 = 1024/4).

However, the upshot is that the read noise relates to the pixel
capacitance which corresponds to well capacity. So absolute
read noise will tend to scale down with pixel size, somewhat
proportionally to the pixel pitch.
Semi-conductor expert bobn2 has written about that in
some posts, e.g. here:
http://forums.dpreview.com/forums/read.asp?forum=1000&message=28533465

So, in this case we can expect 2 electrons of read noise
with the 12Mp sensor, 9 stops of pixel DR (2^9 = 1024/2).

Now, many people, including you-know-who, would still say: "Look at
that pixel-stuffed 12Mp sensor, it lost a stop of DR against the 3Mp sensor,
let's have our 3Mp sensors back!"

But if the 12Mp is downsampled to 3Mp, the (uncorrelated) read noise
will reduce, due to laws of statistics to (2+2+2+2) sqrt(4) = 4. The max
possible signal is 4x1024=4096, and we are back at 10 stops of DR
(10^2 = 4096/4). [reference 1]

OTOH there might be some kind of proportionality (but not
linear) between pixel size and the read noise.

Typically, with current technology, as seen above, it IS linear in pixel pitch.

Secondly, even though the shot noise is averaged out to certain
degree, if the binning is not done on the sensor itself before the
A/D converter, potentially more noise is introduced to the equation.

bobn2 again:

"Downstram amplifier noise is not dependent on pixel size. However, if an area in a final image is made by integrating the contribution of a larger number of pixels (i.e. a higher pixel density sensor) the effect will be to smooth the amplifier noise contributions, resulting in lower noise for the higher pixel density sensor."

Fill factor is a another aspect, AFAIK achieving higher fill factor
with lower mp is easier. Yes, micro lenses come to mind, but light is
still lost between the lenses.

Some people claim it's easier to make efficient microlenses for small pixels,
which would compensate for this. In the future, big pixel microlenses
might catch up, of course.

So I still don't understand how sensor size affect DR. To me (putting
Bayer aside) one binned pixel should have virtually the same DR as a
whole sensor (not like noise which benefits a lot from averaging).
IOW, as I understand this, DR is related to pixel size, not sensor
size.

From [reference 1] above, we see that binning demonstrates how
pixel DR goes up from the effect of having more pixels (whether
we bin them or just look at them (better)) Our eyes will do the
binning when looking at the high-res print. And having more
pixels will be advantageous for the perceived DR, or "DR per area".

Pixel DR is not what defines the final image.

EDIT: And this also shows why a larger sensor with the same
pixel density will have more "image DR".

And another point: Software binning doesn't improve the image in
any way, on the contrary, it throws resolution away. But it demonstrates
what's there in the data, all the time, so it's a useful concept. Edit

Hmmm, so the 12mp APS-c CMOS is sony but the FF version not? I
thought all Nikon sensor were made at the same place but they
customized one part of the process.

There are allegedly some tell-tale signs/marks when a sensor is made by
Sony, if it is scrutinised under a microscope, and these marks are lacking on
the D3 sensor.

Just my two oere
Erik from Sweden

 

[Raist3d]

believing the e-300 has more DR then the e-330. That's not my finding
at all. and I have tested both side to side with the limits of their
exposure at the same ISO checking the extremes (like I did when I
tested the e-330 vs the e-410).

Actually, the test results show the E-330 has more DR than the E-330

• where it counts. But it's a very small difference.

I agree it's small.

My experience reflects what I see with Imatest quite well. The
difference in DR between the E-300/500/330 are marginal at best.

Another reason why I don't trust Imatest one bit.

Keep in mind that I'm testing using dcraw as the standard processing
engine. No noise reduction. No sharpening.

Frankly, I think the Imatest results I've posted are better
controlled than anything I've ever seen you post. But maybe I missed
a post somewhere.

Yeah, looks like you did. Though I didn't use dcraw, so maybe that's my problem. But I sure trust my own assessment of highlights and exposure over a program that determines what noise is supposed to be any day.

dcraw could certainly be my problem. That said in pragmatic terms, using dcraw has its own set of issues, so in a real world scenario, it can't always be counted in as an equivalent solution.

--
Jay Turberville
http://www.jayandwanda.com

--
Raist3d (Photographer & Tools/Systems/Gui Games Developer)
Andreas Feininger (1906-1999) 'Photographers — idiots, of which there are
so many — say, “Oh, if only I had a Nikon or a Leica, I could make great
photographs.” That’s the dumbest thing I ever heard in my life. It’s
nothing but a matter of seeing, and thinking, and interest. That’s what
makes a good photograph.'

 

[Jay Turberville]

I'll side with Jay on the importance of competitive price. The G1
offers no advantages over low-end DSLRs and still has sensor
disadvantages vs APS DSLRs.

Actually, the G1 offers at least four distinct advantages over low-end DLSRs.

1) Size and weight
2) Larger bright viewfinder
3) Tilt Swivel LCD
4) Quieter and lower vibration operation.

There is no way that it will be
successful at the rumored $1200 for a two-lens kit.

Well, they announced $800 for a one lens kit, and that seems OK to me. That means it will probably soon sell for around $700.

--
Jay Turberville
http://www.jayandwanda.com

 

[Steve82]

I'll side with Jay on the importance of competitive price. The G1
offers no advantages over low-end DSLRs and still has sensor
disadvantages vs APS DSLRs.

Actually, the G1 offers at least four distinct advantages over
low-end DLSRs.

1) Size and weight
2) Larger bright viewfinder
3) Tilt Swivel LCD
4) Quieter and lower vibration operation.

The size and weight advantage of the body is minimal at best. (I'm not sure that size is a big deal once a camera is too large for a coat pocket.) I compared the G1 and the K2000 dimensions given here at DPR at the G1 has a mazimum of 5 mm advantage in any one dimension. I haven't really compared the lenses to see how their size compares to other two-lens kits.

The larger viewfinder will only be of practical use if the EVF lag is low enough...or if it doesn't matter for what you are shooting. I remain skeptical about the EVF lag...but we need real reports before drawing conclusions.

The tilt/swivel LCD, combined with a useable LV would prove useful for certain types of photographs. However, these have not drawn huge numbers of people to 4/3 before, so I'm not sure that they will now.

I'm not sure where quieter operation combined with poorer high ISO performance of the 4/3 sensor would prove useful. It might be for some people, but I just don't know who they are.

I just reread what I wrote and it sounds like I am trying to argue about your list of advantages. That was not my intention....I just wanted to point out that the advantages only apply in relatively limited situations. If they are advantages to you, then that is all that really matters. The open question is whether or not there are enough advantages to enough people to justify the premium that Panny is looking to charge.

 

[Guido FORRIER]

hello steve,

i already ordered a body + 14-45 lens for 699 € = cheaper than a new lens for my dsrl.

The evf (i had the camera in my hands at the photokina) is very nice and there no visible lag.

The tilt/swivel screen is very , very usefull to me : plenty of situations where i will use it.I have a painfull back and it is better to tilt/swivel the screen than my back/neck (shooting the ceilings of many castles/buildings ; small children from below etc...)

In my filmday's i used very much the rolleiflex : you can compare the G1 with the rollei , except that in low light situations life view + screen = brighter
and so on ......
Guido

 

[frosti7]

The size and weight advantage of the body is minimal at best.

While on first look it can appear to be so, but if you consider the total bulk(volume) of the camera when compared to the smallest DSLR camera of today the E420, the G1 is still 25% smaller then the e420, plus the G system lens are easily around 50% smaller, add all that to a bag, and you'll receive a much smaller&lighter package and much more convenient setup to carry around with you, I belive that the very handling of "getting it out from a bag and getting it back in" would be much easier because of the small size, changing lens would be easier for this reason and also from the lesser wieght of the lens.

Second, i dont belive that many E420 would want to upgrade to G1, but if a canon 450D user would want to upgrade then the g1 is 40% smaller! and thats significant!

The larger viewfinder will only be of practical use if the EVF lag is
low enough

From the first reports the lag is minimal at worst, after all its a 180 fps EVF,

ofcurse it gets worse at low light, but people tend forget that its also hard to see thru OVF when its dark.

The G1 EVF is considred the best of its class, while its not OVF, it comes with natural advantages that OVF cannot do, e.g: 100% coverage, exposure compensation-real time simulation(a biggie), shutter speed effect simulation(a huge biggie), 50% larger display then the olympus entry level DSLR, much better manual focus ability due to magnification, and much more....

yes it would look less natrual, but i belive that the end effect is more important, and the EVF does have its own advantages.

The tilt/swivel LCD, combined with a useable LV would prove useful
for certain types of photographs. However, these have not drawn
huge numbers of people to 4/3 before, so I'm not sure that they will
now.

Tilt/swivel is really great feature, but i dont think its such a deal braker that people would move to 4/3 system, after all i belive the 4/3 hasn't been a success for pana/oly, hopefully their success lies in m4/3

poorer high ISO performance of the 4/3 sensor

The new mos sensor supposed to have better ISO performance, but even it wont, the ISO of 4/3 while not as good as canon/nikon is still very good and worlds better then compact cameras, i dont think its an issue

(judging form the early samples of G1 the ISO is actually improved from the L10's sensor)

--
Ross
http://www.photoforum.ru/11012

 

[Jay Turberville]

1) Size and weight
2) Larger bright viewfinder
3) Tilt Swivel LCD
4) Quieter and lower vibration operation.

We can debate the relative importance and merits of the advantages all day long. But the fact remains that these are advantages. To say there are no advantages is just plain wrong.

The size and weight advantage of the body is minimal at best.

Compared to what? Most DSLRs are much bigger.

(I'm
not sure that size is a big deal once a camera is too large for a
coat pocket.) I compared the G1 and the K2000 dimensions given here
at DPR at the G1 has a mazimum of 5 mm advantage in any one
dimension.

Huh? Recheck. It is almost a full inch thinner and the K2000 takes up about 60% more space

K2000 123 x 92 x 68 mm 525 g
G1 124 x 84 x 45 mm 360 g

The K2000 weighs about 45% more and the G1 lenses are smaller and lighter.

I haven't really compared the lenses to see how their
size compares to other two-lens kits.

When you do, pay more attention than you did when comparing the camera bodies.

The larger viewfinder will only be of practical use if the EVF lag is
low enough...or if it doesn't matter for what you are shooting. I
remain skeptical about the EVF lag...but we need real reports before
drawing conclusions.

We've had two previews. The lag isn't a big deal. The big bright screen is. Right now there's no reason to think the screen lag will be a problem for the typical user. And the big screen will probably be a big attraction when compared side-by-side in the store.

The tilt/swivel LCD, combined with a useable LV would prove useful
for certain types of photographs. However, these have not drawn
huge numbers of people to 4/3 before, so I'm not sure that they will
now.

There are only two 4/3" DSLRs to ever have this feature. Both have been priced significantly higher than the G1. But regardless, it's a feature that some people want. Hence it is a feature advantage.

I'm not sure where quieter operation combined with poorer high ISO
performance of the 4/3 sensor would prove useful. It might be for
some people, but I just don't know who they are.

Outdoor wildlife photography would be one. But seriously, when is extra noise and vibration an advantage? Quiter and less vibration is pretty much a plus unless you are trying to use your camera to get attention. In that case, the G1 is probably not the right camera to pick.

I just reread what I wrote and it sounds like I am trying to argue
about your list of advantages. That was not my intention....I just
wanted to point out that the advantages only apply in relatively
limited situations.

Well, the way I see it you took a very narrow view in your analysis and were sloppy in your facts. When I see that, I'm inclined to infer an inherent bias.

But anyway, as I said at the outset, my point was that the system has advantages. The original statement that said otherwise is false.

If they are advantages to you, then that is all
that really matters. The open question is whether or not there are
enough advantages to enough people to justify the premium that Panny
is looking to charge.

Sure. But that isn't what I was responding to. I was making the point that the G1 certainly does offer advantages over other cameras.

--
Jay Turberville
http://www.jayandwanda.com

 

[Steve82]

(I'm
not sure that size is a big deal once a camera is too large for a
coat pocket.) I compared the G1 and the K2000 dimensions given here
at DPR at the G1 has a mazimum of 5 mm advantage in any one
dimension.

Huh? Recheck. It is almost a full inch thinner and the K2000 takes
up about 60% more space

K2000 123 x 92 x 68 mm 525 g
G1 124 x 84 x 45 mm 360 g

K2000
122.5 x 91.5 x 73



G1
123 x 86 x 66



I don't consider those significant differences

 

[Steve82]

(I'm
not sure that size is a big deal once a camera is too large for a
coat pocket.) I compared the G1 and the K2000 dimensions given here
at DPR at the G1 has a mazimum of 5 mm advantage in any one
dimension.

Huh? Recheck. It is almost a full inch thinner and the K2000 takes
up about 60% more space

K2000 123 x 92 x 68 mm 525 g
G1 124 x 84 x 45 mm 360 g

K2000
122.5 x 91.5 x 73
http://a.img-dpreview.com/previews/PentaxKM/images/allroundview.jpg

G1
123 x 86 x 66
http://a.img-dpreview.com/previews/PanasonicG1/Images/allroundview.jpg

I don't consider those significant differences

 

[Steve82]

We've had two previews. The lag isn't a big deal. The big bright
screen is. Right now there's no reason to think the screen lag will
be a problem for the typical user.

I haven't seen any of the previews talk about tracking moving objects. I'm not interested in birds, but I am interested in my kids...volleyball, soccer, basketball, etc. Note that I didn't say that the EVF lag was unacceptable, just that I'm skeptical. I just got rid of EVF lag and have no interest in accpeting a noticeable delay.

But regardless, it's (tilt/swivel) a
feature that some people want. Hence it is a feature advantage.

It's a difference. It's not an advantage unless people are willing to pay for it.

But seriously, when is
extra noise and vibration an advantage?

When is poorer performance at high ISOs an advantage? People usually talk about wanting quieter cameras in churches, museums, plays, concerts etc.....which all generally have poor lighting. How many people are going to swap quiet for ISO performance? Some?...probably? Many?....doubtful.

And the comment about wildlife photography...are you kidding? What are people using today?

Well, the way I see it you took a very narrow view in your analysis
and were sloppy in your facts.

My comments were brief, but the analysis was not narrow. m4/3 could potentially offer a camera with compelling reasons to buy into that system. But the G1 at $1200 for a two-lens kit is not it.

 

[Joesiv]

K2000
122.5 x 91.5 x 73
http://a.img-dpreview.com/previews/PentaxKM/images/allroundview.jpg

G1
123 x 86 x 66
http://a.img-dpreview.com/previews/PanasonicG1/Images/allroundview.jpg

I don't consider those significant differences

If you look at the thickness, notice how on the pentax due to the flange back distance for the mirror (and of course the mount parameters) There is a large extruding section forward from the main part of the body. Fourthirds bodies like the e-420/520, have this as well, it's required if you have a mirror, unless you want a generally thick body like the e-330 or L1.

The G1 doesn't have this since the sensor can be very close to the lens. What this means is that measurements of the body that include the grip, will hide the fact that the lens will sit much closer the the rear of the body. With lens mounted, this will translate to even smaller dimensions if you facter in the lens.

I think it's something easily missed, but is one of my revelations when considering the e-420 for portability.

If you had a pancake lens on the G1, the lens may not even stick out farther than the trip, that's pretty significant!

--
Cloverdale, B.C., Canada
Olympus e-510 L1
http://joesiv.smugmug.com

 

[John Carson]

When is poorer performance at high ISOs an advantage? People
usually talk about wanting quieter cameras in churches, museums,
plays, concerts etc.....which all generally have poor lighting. How
many people are going to swap quiet for ISO performance?
Some?...probably? Many?....doubtful.

4/3 DSLRs have the same high ISO disadvantage. It is true that Canikon have an advantage over Olympus/Panasonic in the high ISO stakes but this isn't something that just came about with the introduction of EVIL cameras.

How many people are going to swap quiet for ISO performance? Well, if you are in a situation where you aren't allowed to make the noise of a DSLR, then it comes down to a swap between a picture and no picture. With museums, you are generally photographing static objects and ISO 800 will generally be enough. With plays and many concerts etc., it is at least bad manners and it may be forbidden to use a DSLR.

I am not disputing the advantages of good high ISO performance, just suggesting that more may be willing to make the tradeoff than you think.

--
john carson

 

[xavozin]

just to note here, the weight of the G1 is 385g (not 360) body only without anything, with the battery and lens it is 640g. the only DSLR that is close is E420 but you have to give up image stabilization, the hand grip and live with an OVF that is simply too small when compared to the G1's EVF. given all these advantages of G1, i find the comparison with E420 ridiculous. if size is so important and one is willing to give up features to have the lightest possible package, then why not give up large sensor + interch. lenses and get an ultra compact P&S.

the removal of the reflex and OVF has many consequences like an easy to use LV mode, less complex design and better optical performance from wide angle lenses, no mirror slap or shake, no light to enter through the OVF during long exposure, you see a digital preview before shooting that is much closer to the digital image u can capture than what u would see in the OVF (the OVF could be nice if the device was something like a binocular not a camera). whoever appreciates all these advantages will not care about how small or big the size difference is (between G1 & DSLRs).

(I'm
not sure that size is a big deal once a camera is too large for a
coat pocket.) I compared the G1 and the K2000 dimensions given here
at DPR at the G1 has a mazimum of 5 mm advantage in any one
dimension.

Huh? Recheck. It is almost a full inch thinner and the K2000 takes
up about 60% more space

K2000 123 x 92 x 68 mm 525 g
G1 124 x 84 x 45 mm 360 g

K2000
122.5 x 91.5 x 73
http://a.img-dpreview.com/previews/PentaxKM/images/allroundview.jpg

G1
123 x 86 x 66
http://a.img-dpreview.com/previews/PanasonicG1/Images/allroundview.jpg

I don't consider those significant differences

--
Xavo

image is the only validation and most of their life isn't real.
modified from Sam Sparro - Black and Gold.

 

[viztyger]

just to note here, the weight of the G1 is 385g (not 360) body only
without anything, with the battery and lens it is 640g. the only DSLR
that is close is E420 but you have to give up image stabilization,
the hand grip and live with an OVF that is simply too small when
compared to the G1's EVF. given all these advantages of G1, i find
the comparison with E420 ridiculous. if size is so important and one
is willing to give up features to have the lightest possible package,
then why not give up large sensor + interch. lenses and get an ultra
compact P&S.

Well put. The size of the µ4/3 mock-up demonstrates your point: it's smaller but does away with the EVF, articulated screen and a zoom lens. And who knows what else.

the removal of the reflex and OVF has many consequences like an easy
to use LV mode, less complex design and better optical performance
from wide angle lenses, no mirror slap or shake, no light to enter
through the OVF during long exposure, you see a digital preview
before shooting that is much closer to the digital image u can
capture than what u would see in the OVF (the OVF could be nice if
the device was something like a binocular not a camera). whoever
appreciates all these advantages will not care about how small or big
the size difference is (between G1 & DSLRs).

To a degree; compactness is one of the benefits of the µ4/3 system in addition to the ones you mention above. Given all those benefits and features, the camera is at the same time very compact and a much more ideal travel camera than my heavy D300.

I purchased a D300 due to its 100% coverage viewfinder, among other things. I often have to carefully consider the edges of the frame, especially when framing architecture shots. Might as well compare the size of the G1 to DSLRs featuring 100% viewfinders...

--
Björn

galleries: http://www.pbase.com/viztyger

 

[GregGory]

Sorry Erik, I forgot about this thread...

I re-read it and can't find any errors in what I wrote, but
"everything being
equal" is of course not clear-cut, so is open for interpretation.

There are two separate issues:

1) How to compare fairly if the Mp count differs between two cameras.
Pixel DR is not fair.

How many pixel does a measurement have to span to yield a fair measure in your opinion? The whole picture, or just a 'bunch' of pixels?

2) What constitutes everything else being equal, or how does the
read noise, FWC etc. vary with pixel size.

There seems to be a theoretical, a common belief, and an empirical line.

However, the upshot is that the read noise relates to the pixel
capacitance which corresponds to well capacity. So absolute
read noise will tend to scale down with pixel size, somewhat
proportionally to the pixel pitch.

This is the 'theoretical' line. I'm not sure the empirical evidence supports it, OR there are other variables that have to be singled out. More specifically, I would like to see the small sensor vs. stuffed small sensor comparison before I'd really believe it.

Semi-conductor expert bobn2 has written about that in
some posts, e.g. here:
http://forums.dpreview.com/forums/read.asp?forum=1000&message=28533465

So, in this case we can expect 2 electrons of read noise
with the 12Mp sensor, 9 stops of pixel DR (2^9 = 1024/2).

Now, many people, including you-know-who, would still say: "Look at
that pixel-stuffed 12Mp sensor, it lost a stop of DR against the 3Mp
sensor,
let's have our 3Mp sensors back!"

But if the 12Mp is downsampled to 3Mp, the (uncorrelated) read noise
will reduce, due to laws of statistics to (2+2+2+2) sqrt(4) = 4. The max
possible signal is 4x1024=4096, and we are back at 10 stops of DR
(10^2 = 4096/4). [reference 1]

OTOH there might be some kind of proportionality (but not
linear) between pixel size and the read noise.

Typically, with current technology, as seen above, it IS linear in
pixel pitch.

Hmmm, the math looks good, at least I cannot argue against it! So theoretically, 4 x 9 stops equals 1 x 10 stop. The question remains how precise the "4 and 2 electron noise estimate" really is.

Secondly, even though the shot noise is averaged out to certain
degree, if the binning is not done on the sensor itself before the
A/D converter, potentially more noise is introduced to the equation.

bobn2 again:
"Downstram amplifier noise is not dependent on pixel size. However,
if an area in a final image is made by integrating the contribution
of a larger number of pixels (i.e. a higher pixel density sensor) the
effect will be to smooth the amplifier noise contributions, resulting
in lower noise for the higher pixel density sensor."

So this is what the upcoming Fuji sensor is all about in high-iso mode?

Fill factor is a another aspect, AFAIK achieving higher fill factor
with lower mp is easier. Yes, micro lenses come to mind, but light is
still lost between the lenses.

Some people claim it's easier to make efficient microlenses for small
pixels,
which would compensate for this. In the future, big pixel microlenses
might catch up, of course.

I've read the opposite, but it was several years ago. It's expensive at least, and small cheap compacts cannot 'afford' to spend too much on this. But if it really is true, this might have the opposite implication on large sensors, An argument for more pixels.

So I still don't understand how sensor size affect DR. To me (putting
Bayer aside) one binned pixel should have virtually the same DR as a
whole sensor (not like noise which benefits a lot from averaging).
IOW, as I understand this, DR is related to pixel size, not sensor
size.

From [reference 1] above, we see that binning demonstrates how
pixel DR goes up from the effect of having more pixels (whether
we bin them or just look at them (better)) Our eyes will do the
binning when looking at the high-res print. And having more
pixels will be advantageous for the perceived DR, or "DR per area".

Pixel DR is not what defines the final image.

EDIT: And this also shows why a larger sensor with the same
pixel density will have more "image DR".

I'm not following, 'across' the image, or locally?

And another point: Software binning doesn't improve the image in
any way, on the contrary, it throws resolution away. But it demonstrates
what's there in the data, all the time, so it's a useful concept. Edit

I understand your stand, that DPR and the general crusaders against pixel stuffing should provide some evidence for their case, but OTOH they point to the empirical findings, that among DSLRs and among small sensors, more pixels seem to generate more noise. I think a thorough test between the FX150 and LX3 will bring a lot to the table.

 

[Ehrik]

Sorry Erik, I forgot about this thread...

Happens to me too all the time. You can see which threads you've been
posting in (Kudos Phil!), but not list only those where you've got replies
you haven't read, and when you post in many, it's easy to miss some.

Anyway, no need to be sorry, I enjoy discussing this.

1) How to compare fairly if the Mp count differs between two cameras.
Pixel DR is not fair.

How many pixel does a measurement have to span to yield a fair
measure in your opinion? The whole picture, or just a 'bunch' of
pixels?

Equal proportions of the full image, either the whole image or 5% or whatever.

how does the read noise, FWC etc. vary with pixel size.

There seems to be a theoretical, a common belief, and an empirical line.

Sort of. There's a common belief because sites like DPR have the possibility
to propagate their (pre)conceptions to many people from a seemingly
authoritative position.

There are empirical tests, done by a few independent persons that seem
to point in about the same direction. Then bobn2 seems to be the only one
(that I know of) of those who post on these forums that has presented a
theory that fits the empirical findings.

More specifically, I would like to see the small sensor vs.
stuffed small sensor comparison before I'd really believe it.

So would I, very much.

There must be a point when the pixels are too small. But
the FZ50 comparing well to the D3 and 1Ds3 per area
suggests that at 2µm there is still no reason to worry. And the
LX3 seems (subjectively, but not yet conclusively) to do even
better with its 2µm pixels.

Still, those who think pixel stuffing is terrible were complaining already when
the FZ30 with its 8Mp came. It's time they "put up or shut up".

For CMOS there's the potential to cancel reset noise which gives low high-ISO
noise (e.g. Canon's DSLRs and the latest Nikons) and would give lower noise
at base ISO if they found a better way to get the data off the sensor. But
CMOS need more transistors per pixel, so may be harder to shrink. Canon didn't
promise IQ improvements by making small CMOS sensors for compacts.

Hmmm, the math looks good, at least I cannot argue against it! So
theoretically, 4 x 9 stops equals 1 x 10 stop.

Yes! And then we have still disfavoured the 4 x 9 by throwing out a lot
of resolution. In the end it's our eyes that are the judges, and the limit of
DR (once we've exposed properly to the right) is how much detail
we can make out in the shadows. Noise is less disturbing when it
has to compete with image detail.

And a sophisticated NR program can be more effective using the higher
resolution (prior to downsizing if we need to downsize).

Of course if we split the 1x10 pixel into four new pixels, they won't
have 9 stops of DR, but 10. This also points to the inherent difficulty
in comparing DR between sensors of different pixel counts.

Noise has to be seen as not only having an amplitude (z direction) but
an extension in the sensor plane (x and y direction) relative to the
image and the subject detail.

The question remains
how precise the "4 and 2 electron noise estimate" really is.

Yes. It would be nice if the camera manufacturers would declare the
read noise at various ISOs, the "speeds" of the ISOs relative to a standard,
full well capacity and the quantum efficiency of their cameras/sensors.

So this is what the upcoming Fuji sensor is all about in high-iso mode?

I can't say, haven't studied that one much.

Some people claim it's easier to make efficient microlenses for small
pixels

I've read the opposite, but it was several years ago.

Well at least intuitively you'd think it would be easier to make big ones
because the precision demands would be less tight (in absolute measures).
The big-pixel microlenses would have to be higher of course, a thicker layer,
so maybe that's where the difficulty enters.

Another factor mentioned sometimes is how digicam lenses can come close
to the sensor, making near-telecentric designs easier. And be tailored to
a fixed lens. Whereas DSLR sensors, made to be compatible with a
variety of legacy glass that's not so tele centric, may have to make
compromises with the microlenses. I don't know if there is anything in this.

Our eyes will do the
binning when looking at the high-res print. And having more
pixels will be advantageous for the perceived DR, or "DR per area".

Pixel DR is not what defines the final image.

EDIT: And this also shows why a larger sensor with the same
pixel density will have more "image DR".

I'm not following, 'across' the image, or locally?

Well, with "image DR" I'm trying to say "normalised for the pixel count".
As opposed to "pixel DR", the 100% view.

I understand your stand, that DPR and the general crusaders against
pixel stuffing should provide some evidence for their case, but OTOH
they point to the empirical findings, that among DSLRs and among
small sensors, more pixels seem to generate more noise.

They point to them generating more noise per pixel . They don't
properly normalise to a common viewing size.

And for DR it's even worse, they point to factors like blown highlights
that are decided by tone curves and metering and then imply that it
says something about sensors and pixel density. Also, see
http://forums.dpreview.com/forums/read.asp?forum=1000&message=29616791

a thorough test between the FX150 and LX3 will bring a lot to the table.

Absolutely. But they did mention improved electronics
in the press release for the LX3 and not (?) the FX150
so if the LX3 were better per area, it would still not be conclusive.
But it would be one of the more useful tests to do on recent small-pixel
cameras.

Just my two oere
Erik from Sweden