Steen Bay
Veteran Member
And the microlenses probably are highly optimized to give us the highest possible QE at f/2.8 and above, but Canon forgot to mention in the manual that it isn't the best idea to use e.g. 7D with the 50/1.2 wide open.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Light loss on current CMOS sensors at big apertures
- Thread starter Bogdan_M
- Start date
G Cogger
Active member
In terms of light gathering, 0.3 EV loss actually makes it the equivalent of f/1.55, so it's still worth having the lens over, say, an f/1.8. Yes, it's not ideal, but that's life - hardly worth getting upset about
It's good to know, though. My f/1.4 lenses are old manual lenses, so I'll know to keep an eye on exposure now.Midnighter
Veteran Member
How about this:
They brought out Premium Guinness, four to ten times more expensive than Regular. But you liked the superior taste and fronted the money. You never knew it had a fly in it until someone pointed out its left wing, stuck to your lip. Then some people who drank Regular told you it was a good thing, because the fly was necesary to give Premium a proper taste.... they then went back to happily drinking their Regulars, some laughing under their breath and others wondering what the fuss was about.
They brought out Premium Guinness, four to ten times more expensive than Regular. But you liked the superior taste and fronted the money. You never knew it had a fly in it until someone pointed out its left wing, stuck to your lip. Then some people who drank Regular told you it was a good thing, because the fly was necesary to give Premium a proper taste.... they then went back to happily drinking their Regulars, some laughing under their breath and others wondering what the fuss was about.
jerome_munich
Leading Member
There is certainly an amount of pixel shading for some lenses, but it does not depend on aperture only . Hence the theory of luminous-landscape is not correct. There is no way the manufacturer could correct for a given aperture (say: f/1.4) because the problem will not be the same for all f/1.4 lenses.
The shading depends on the angles of the rays entering the sensor, which in turn depends on the aperture (indeed) but also of the distance between the exit pupil and the sensor. For a telecentric lens, which is the extreme case, the exit pupil is at the infinite and the rays are all orthogonal to the sensor: no shading at any aperture.
You can (roughly) estimate the distance of the exit pupil by unscrewing a lens and examining the film side. You will see the diaphragm behind a part of the optics. On a telecentric lens, the diaphragm appears to be very far away. It looks a bit like looking through the wrong end of binoculars.
Now on this telecentric lens, the diaphragm appears far away and small (but very bright if it is a fast lens). Even if you would be at the bottom of a well (as the pixels are on a ccd), you would still see it all. OTOH, if the diaphragm would be huge (bigger than the well hole) and close, the further down you go, the less of it you see.
The shading depends on the angles of the rays entering the sensor, which in turn depends on the aperture (indeed) but also of the distance between the exit pupil and the sensor. For a telecentric lens, which is the extreme case, the exit pupil is at the infinite and the rays are all orthogonal to the sensor: no shading at any aperture.
You can (roughly) estimate the distance of the exit pupil by unscrewing a lens and examining the film side. You will see the diaphragm behind a part of the optics. On a telecentric lens, the diaphragm appears to be very far away. It looks a bit like looking through the wrong end of binoculars.
Now on this telecentric lens, the diaphragm appears far away and small (but very bright if it is a fast lens). Even if you would be at the bottom of a well (as the pixels are on a ccd), you would still see it all. OTOH, if the diaphragm would be huge (bigger than the well hole) and close, the further down you go, the less of it you see.
Bogdan_M
Veteran Member
Andy Westlake wrote:
Hello Andy, thanks for your answer and for the interesting quantitative data.
Two questions come to mind : is this pretty good fstop scaling a result of Panasonic anticipâting that mFT sensors will be paired with scary fast lenses sometime ijn the future, so the microlenses were designed as such? This actually makes the GF1 look pretty good.
I saw that the Noct f 0.95 was introduced after the M8, so I can only wonder: how telecentric or near telecentric is the design of this lens? This might have something to do with the issue too, or am I misunderstanding? A near telecentric lens should have less pixel vignetting issues, regardless of the sensor, or is this affecting only sensor periphery vignetting?
http://www.flickr.com/photos/bogdanmoisuc/
Hello Andy, thanks for your answer and for the interesting quantitative data.
Well it seems for some sensors the returns diminsh quite a lot. In fact, for somebody interested in low light/thin dof photography, it may be a viable option to dump the camera and go for the cheaper lesn + FF, which would give more "real" DOF and low light capabilities, with the absence of increasing noise (by staying below f 2).
That's a great result, and it somehow reassures me towards the mFT sensors (more specifically, their microlenses). It seems here the effect is quite small, acceptable until the widest aperture.Here's a little quick'n'dirty experiment I did about 6 months ago investigating the pixel vignetting phenomenon using the Leica Moctilux 50/0.95 on the GF1. This is aimed at looking at DoF and background blur effects, by measuring the blur circle from an out-of-focus point light source (here an LED flashlight about 4m from the camera in a darkened room) in the centre of the frame across a range of apertures. Clearly, the blur circle diameter should be directly proportional to the aperture diameter. Here's the results:
What you see immediately from this is that there's no sharp cutoff in blur circle at a specific aperture - however a little further calculation shows that you don't get the full theoretical benefit of the maximum aperture either. In this I'm assuming pixel vignetting affects are negligible at F2 (which, given the results, seems fair), and calculating what the blur circle diameter should be at larger apertures.
This shows that even at F1.2 there's not a huge effect - maybe 1/6 stop loss. At F0.95, though, the blur circle is considerably smaller than expected, corresponding to an apparent aperture of about F1.1, i.e. maybe a half stop loss in light. However, even then you do still get a bit more blur than when the lens is set to F1.2, so the F0.95 lens isn't completely wasted. (However it does reinforce the idea that the Voigtlander Nokton 50/1.1 may just be better value than the Noctilux, as if we didn't know that all along.)
Two questions come to mind : is this pretty good fstop scaling a result of Panasonic anticipâting that mFT sensors will be paired with scary fast lenses sometime ijn the future, so the microlenses were designed as such? This actually makes the GF1 look pretty good.
I saw that the Noct f 0.95 was introduced after the M8, so I can only wonder: how telecentric or near telecentric is the design of this lens? This might have something to do with the issue too, or am I misunderstanding? A near telecentric lens should have less pixel vignetting issues, regardless of the sensor, or is this affecting only sensor periphery vignetting?
--
http://www.flickr.com/photos/bogdanmoisuc/
Bogdan_M
Veteran Member
Great post, I had jjust asked this question in another reply. Maybe the differences in telecentricity are not that big between manufacturer's lenses at the same focal length? (I doubt that Canon or Nikon is anxious to fix this issue on Sigma lenses, but they would want their own to work well).
Now the only issue is getting exit pupil distance for all those fast lenses and the 2D graph at LL would be much more meaningful : on x axis we would plot lens lens distance pupil nd on the Y axis Tstop loss (minus the Tstop resulting from absorbtion by the lmens elements themselves). That would also give a better assessment of the sensor performance in itself, regarding this issue.
--
http://www.flickr.com/photos/bogdanmoisuc/
knickerhawk
Veteran Member
The difference is attributable in large part to pixel pitch. In fact, the phenomenon has to be directly related to the size of the sensel opening (like the difference between shooting basketballs at a bigger rim vs smaller rim - more will go through the bigger rim).
The fact that a pixel pitch of approx. 6.3 has such a wide range of results (the 1DsMarkIII at -.35 and the D60 at -.75) says that a lot of the phenomenon is NOT an inherent physical limitation. Good engineering/design can overcome much of it. It makes me optimistic that better microlenses and other tweaks can minimize the effect. And just because Canon claims the 7D has improved microlens design doesn't mean that it optimized the 7D microlens design for this (rather limited) issue.
--
My photos: http://www.pbase.com/imageiseverything/root
How the Resulting EV value is calculated? Some portion of the Image, if so, where?
If it is the entire image, how much light loss correspond to Optical Vignetting and how much to Pixel Vignetting?
There is a big difference shown in the case of 7D and 550D, is this because different microlens or just sample variation? I would expect the 7D to outperform if not match the 550D.
A comparison to film as a reference would have been nice too.
If it is the entire image, how much light loss correspond to Optical Vignetting and how much to Pixel Vignetting?
There is a big difference shown in the case of 7D and 550D, is this because different microlens or just sample variation? I would expect the 7D to outperform if not match the 550D.
A comparison to film as a reference would have been nice too.
Marc Dubovoy clarified in a forum thread at LL that the data are from the image center.
The smaller the sensel, the longer the tube compared to its opening diameter (ie, when sensels get smaller they do not get shallower by the same amount). You see for all the results that the observed effect correlates reasonably well with the sensel size. But microlenses certainly also have their effect (they try to compensate for this effect).
That should have been in the article. But thanks for the answer.
But the 7D and 550D share the same pixel pitch Both have sensors APS-C 18MP.
jerome_munich
Leading Member
Well... First of all, the issue only really concerns fast wide-angles and there are huge differences as to the position of the exit pupil from lens to lens. Older designs in particular usually have their exit pupil closer to the sensor plane.
Secondly, it would theoretically be possible for some manufacturer to recognize the exact lens at to have a correction table in firmware. But that would fail for unknown lenses (either third party or old lenses without electronic i.d.). Maybe they indeed do that, but should we blame them?
Thirdly, the issue was about the sensor limiting the maximum usable aperture. It does not. You just need telecentric lenses.
Except that we don't know which lens they used for that graph. But if someone is interested, I can probably measure Sony/Minolta lenses fairly easily.
kenw
Veteran Member
Great data, thanks for sharing!
Since the pixel vignetting is a function of the exit pupil distance it will be interesting to see what happens with the 25/0.95 where the exit pupil is likely closer to the sensor. Might see a more significant effect starting at smaller apertures in that case.
--
Ken W
Rebel XT, XTi, Pany G1, LX3, FZ28, Fuji F30, and a lot of 35mm and 4x5 sitting in the closet...
Since the pixel vignetting is a function of the exit pupil distance it will be interesting to see what happens with the 25/0.95 where the exit pupil is likely closer to the sensor. Might see a more significant effect starting at smaller apertures in that case.
--
Ken W
Rebel XT, XTi, Pany G1, LX3, FZ28, Fuji F30, and a lot of 35mm and 4x5 sitting in the closet...
Andrew Westlake
Veteran Member
I don't. What's more, I don't think it really matters. The lens may be a little slower than specified, the aperture may not be 100% accurate, but that doesn't affect the take-home message - that pixel vignetting doesn't seem to be a huge problem with this combination even at F1.2.
--
Andy Westlake
dpreview.com
bobn2
Veteran Member
Ho bogdan,
the problem is the at fill factor just says that the microlenses are as big as they can go for that pixel size. Small pixels give small ML diameter, if the metal stack height on the chip doesn't shrink accoordingly you can't reduce the FL of the microlens and its f-number rises accordingly. According to chipworks, at the time of the 7D Canon hadn't updated its process for a while. I think newer sensors are on its new line, which might have thinner metal.
--
Bob
Bogdan_M
Veteran Member
A-ha
. Thank you for the clarification Bob. That makes sense, I imagine other areas will receive improvements as well. At present, Canon seems to be well behind Nikon in DR and high ISO noise.Looking forward for the new sensors from Canon.
--
http://www.flickr.com/photos/bogdanmoisuc/
bobn2
Veteran Member
Thank, Daniel, really interesting. It just goes to show just how very careful one needs to be trying to divine things like sensor performance from raw data. The 'brain damage' can entirely disrupt these results. It strikes me that we don't know the f-numbers at which DxO does its measurements.
On the gain adjustment, I think the pattern I found in 7D read noise is interesting
I looks like Canon is working with 'primary' ISO's of 100,400, 1600 andswitching the 'secondary' amplification to achieve the intermediates. My suspicion as to why they're doing this is to lower the read noise at 200 and get it more in line with the Sony sensors.
--
Bob
bobn2
Veteran Member
It depends, for the D3S, yes for the D3100, not so sure. The D3100 is about as good as the 7D, a bit behind that 550D and 60D (which seem to have a different and better sensor than the 7D) . The advantage Nikon has being fabless is that they can shop around for a new process when they need it - I think that's what happened between the D3 and D3S. I think also the effect we're talkng about in this thread might be why the D3S reputedly performs even batter in low light than the bare figures suggest. Low light means small f-numbers and I suspect that the D£S has ver fast microlenses.A-ha
The 1DIV, 550D and 60D are at the top of current sensor performance (although no-one's testing microlens speed at present)
--
Bob
Keyboard shortcuts
- f
- Forum