Bogdan_M
Senior Member
Thank you, thread bookmarked. It will digested slowly and thoroughly
--
http://www.flickr.com/photos/bogdanmoisuc/
-
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
--
http://www.flickr.com/photos/bogdanmoisuc/
Steen Bay
Veteran Member
"Camera makers 'game the system' increasing gain (ISO) to recover t-stop loss" (DxO, graph #3)
Does that mean, that if I'm shooting in Manual mode with a Canon 7D at iso200, f/1.2 and 1/125, then I'll actually be shooting at iso400, although EXIF says iso200? Well, guess that's what the article actually says, and I find that rather surprising, to say the least! (And guess that BSI sensors and better (faster) microlenses could help solving the problem)
Skipper494
Forum Pro
Despite being a mathematician and engineer, among other disciplines, and being fascinated by obtuse problems, I think we should use our cameras as handy tools to learn how to make better photos and not worry about the details. When I read a good book by electric light, I don't worry about what the electrons are doing.
I have a feeling that, if a fly got into some posters' Guinness, they would make it cough up the last drop.
Skipper.
I have a feeling that, if a fly got into some posters' Guinness, they would make it cough up the last drop.
Skipper.
tko
Forum Pro
Yes, it is an interesting article, but I'm not sure I believe it. The angle in which the rays hit the CCD vary from the center to the edge, right? So the edges would be darker than the center. You couldn't compensate by simply boosting the ISO, you'd have to boost the ISO as a function of the distance from the center. The boost would be lens unique.
Not a trivial task, and one that I don't think is being done today. RAW users would cry bloody murder if their raw files were being altered, if they weren't, extreme falloffs at the edge would be noticed.
Besides, micro lenses are supposed to alleviate this phenomenon.
Lastly, full frame sensors are well, larger, and so the angle of light at the edges are more extreme. They should have more light fall off, not less! See this article for confirmation.
http://www.brisk.org.uk/imatest/falloff2.html
Bottom line is, I think DXO data is questionable at best. If you want to read more, see this articles:
http://www.digitaldarrell.com/DDBlog-ShouldNikonMakeA35mmSizedSensor.asp
http://www.pco.de/fileadmin/user_upload/db/download/pco_cooKe_kb_shading_0603_s.pdf
Not a trivial task, and one that I don't think is being done today. RAW users would cry bloody murder if their raw files were being altered, if they weren't, extreme falloffs at the edge would be noticed.
Besides, micro lenses are supposed to alleviate this phenomenon.
Lastly, full frame sensors are well, larger, and so the angle of light at the edges are more extreme. They should have more light fall off, not less! See this article for confirmation.
http://www.brisk.org.uk/imatest/falloff2.html
Bottom line is, I think DXO data is questionable at best. If you want to read more, see this articles:
http://www.digitaldarrell.com/DDBlog-ShouldNikonMakeA35mmSizedSensor.asp
http://www.pco.de/fileadmin/user_upload/db/download/pco_cooKe_kb_shading_0603_s.pdf
Well, I think the point was not that the fly drank a microliter of our Guinness, but that it's been in the glass all along and no one told us. Just because no one really noticed the fly, much less the missing microliter, is irrelevant- it's the principle of the thing that matters. ; )
"Association is not causation" as they say in the statistics trade. There are lots of other things you could plot on the X-axis, eg, year of sensor design, that might be equally, or more closely associated.
For a start, I don't think there is any association between pixel pitch and light: I think there are two populations, one of 24 x18 sensors, and one of 36 x 24 sensors, for neither of which is there any association between pixel pitch and light loss. Light loss is less for the 36 x 24 sensors and they have, mostly, bigger pixels, which creates a spurious impression that pixel pitch and light loss are associated.
24 x 18 sensors have much worse vignetting of the lens image circle than 36 x 24 sensors.
--
A l'eau, c'est l'heure! (French naval motto)
For a start, I don't think there is any association between pixel pitch and light: I think there are two populations, one of 24 x18 sensors, and one of 36 x 24 sensors, for neither of which is there any association between pixel pitch and light loss. Light loss is less for the 36 x 24 sensors and they have, mostly, bigger pixels, which creates a spurious impression that pixel pitch and light loss are associated.
24 x 18 sensors have much worse vignetting of the lens image circle than 36 x 24 sensors.
--
A l'eau, c'est l'heure! (French naval motto)
Steen Bay
Veteran Member
But the fly actually drank 25-50% of our beer (light/photons) and replaced it with water!
Bogdan_M
Senior Member
You can see the plots. If you notice any other assoiciation, let me know.
So it's not pixel pitch, it's sensor format that correlates with light loss for fast lenses. But that doesn't mean it's avignetting issue.
Yes but here the data are presented for the same lenses, which happen to be FF lenses (36x24 image circle). Sure you are not implying that vignetting is stronger on APS-C than on FF sensors for the same lenses. Once again, that is easy to see, the data is there. So it's not correlated to vignetting, on the contrary.
--
http://www.flickr.com/photos/bogdanmoisuc/
Bogdan_M
Senior Member
This is another problem, and this one is known. Vigneting on digital sensors is fixed by boosting the gain on the peripherral part of the sensor, although it is not lens unique (the point is not to fix the lens vignetting, but to fix the sensor vignetting, so that a picture taken on film or on a digital sensor would show the same amount of vignetting.
You're taliking about vignetting again, not the same issue.
DXO data is data, measured scientifically. I have very little to doubt it. Causal interpretations can be easily false, but I see little reason for a trusteable source to shoot themselves in the foot going against camera makers with false data.
What you seem to suppose is that rays hitting the center of the sensor are perpendicular and only light rays hitting the periphery are oblique. What happens actually is, although there is a difference on average between the angle of the rays, even rays hitting the center sensels can arrive at steep angles, the faster the lens, the steeper the angle.
What seems to be happening is that some of these rays don't make it on the sensor, and it seems that the problem is sensor dependent, there is something in the design of APS-C sensors which results in more light getting wasted this way. The author of the article thinks it's the depth of the sensor "wells", while other posters seem to think it has to do with the aperture (or fstop) of the microlenses.
The complexity of the problem is further increased by the fact that some of the loss of light is surely attributable to the lens itself (glass and coatings absorb and reflect some of the light, not all photons make it through the exit pupil). This is responsible, probably for 0.2 or 0.3 of the Tstop difference, and is independent of the sensor design. Which actually makes the differences between the sensors look even bigger : this means that there is virtually no Tstop loss on the 5D (so you get what you pay for when putting the 50L on the 5D) but there's a big loss on a 7D (more than half the light that you are supposed to gain by going from f 2 to f 1.2 is lost), so you get much less than you pay for.
--
http://www.flickr.com/photos/bogdanmoisuc/
Bogdan_M
Senior Member
Indeed. To put it more intuitively, they talk about sensel vignetting or microlens vignetting at large fstops.
--
http://www.flickr.com/photos/bogdanmoisuc/
But if the hypothesis about pixel wells and the obliquity of light was correct there would be an effect of pixel pitch - that is why they show this graph - and there isn't. LL has simply misinterpreted the data.
--
A l'eau, c'est l'heure! (French naval motto)
--
A l'eau, c'est l'heure! (French naval motto)
Bogdan_M
Senior Member
The graphs read "T-stops loss depends on sensor", it doesn't say that it depends on pixel pitch.
The purpose of the letter is to present the facts and ask for explanations.
I was asking for interpretations, and I would like to see some. If the problem is just the micro lens design, it is surprising that one of the oldest cameras in the bunch has the least loss, while the newest sensors fare worse.
Does this mean that while creating gapless microlenses, the manufacturers actually decreased their Fstop? Or is the problem related to something else?
Does this also mean that as a result, and APS-C sensor paired with fast lenses will show more DOF than film (assuming 35 mm film would be cropped to match exactly the APS-C format?
If you have some ideas, please do share.
--
http://www.flickr.com/photos/bogdanmoisuc/
Yes, it seems probable that a company to whom scientific measurements are a major part of their business-model would spread falsified information. As far as I know there as been no credible dispute of the validity of DxOLabs measurements.
Some numbers are all over the place for cameras that are supposed to have the same or almost identical sensor.
Also as a function of pixel pitch, the numbers don't make sense either.
My understanding is that at larger apertures manufacturers are at worst, cheating and boosting Digitally, the pixels affected by the light fall-off at the corners, which when averaged for all pixels amount to what it seems to upping the nominal gain.
Some of this data seems to be in conflict with their other DxOMark numbers.
DxO uses the DxOmarlk as a PR tool.
I wonder what are they going to do, now that Adobe has implemented Lens Profiling and Correction.
Also as a function of pixel pitch, the numbers don't make sense either.
My understanding is that at larger apertures manufacturers are at worst, cheating and boosting Digitally, the pixels affected by the light fall-off at the corners, which when averaged for all pixels amount to what it seems to upping the nominal gain.
Some of this data seems to be in conflict with their other DxOMark numbers.
DxO uses the DxOmarlk as a PR tool.
I wonder what are they going to do, now that Adobe has implemented Lens Profiling and Correction.
Yes, interesting results. I think this is about uncovering details of the pixel design in each sensor, and have had talks about this with bobn2 above in his previous post about the subject. The pdf paper he refers to is important here, and talks about the details of the pixel design (i.e. microlens F#, stack height, active area of the pixel, pixel size) all of which come into play here. And in order to uncover such details requires meticulous, excellent measurements which I think Dxomark does well.
Interesting that they report on F# = 1.2 and 1.4, I would guess that they would have done more measurements at the higher F#'s, I hope there comes out with more results and details of the measurements, i.e. I'd kinda like to see if and when the light losses level off at higher F#'s for the various sensors.
Also interesting is the general trend with pixel size, but there are also outliers with the Nikon D70 series.
Hopefully can read more about this in the future.
Chris
kenw
Veteran Member
As someone pointed out in the LL forums and Mark confirmed his error, the assertion that this is related to CMOS is completely incorrect. Many of the "worst" performers on his chart are in fact CCD sensors.
Pixel design, CCD or CMOS, and more importantly microlens design can have a measurable impact on the acceptance angle for a pixel and this can impact both vignetting and apparent light loss at very wide apertures. That appears to be what the DXO data is showing.
As a previous poster pointed out I think Mark's idea that the camera manufacturer should not compensate for this is in fact an even less desirable situation. I'd want to be able to assume proper reciprocity behavior in exposure for various aperture and shutter speed combinations. More interesting is that this implies that adapting manual legacy lenses to these sensors will result in a bit of reciprocity failure at wide apertures since the camera is unable to gain compensate without any knowledge of the aperture setting - obviously this would still wouldn't cause an exposure error since most manual users meter stopped down anyway and rarely depend on reciprocity.
In the larger context I believe he's making a mountain out of a mole hill at this point. The 1/3 to 2/3 of a stop is nothing of any concern, and I really haven't met a photographer who buys expensive lenses that are a fraction of a stop faster than another because of the additional speed - rather they are usually going for other qualities of the more expensive lens.
More interesting will be to see if future measurements show any impact on narrow DOF. I suspect that difference will be very small as well, again most people buy the more expensive slightly faster lens because often times the bokeh is just better at all apertures on the more expensive lens.
Anyway, it is an interesting article but I felt it read as if it was written in a rush, not well thought out, and rather half baked at this point. When the focus tests are done and it is properly edited to remove the obvious errors and written more clearly I think it will be a great article. Whether in the end the technical minutia contained will be of any relevance in "real world" photography remains to be seen.
--
Ken W
Rebel XT, XTi, Pany G1, LX3, FZ28, Fuji F30, and a lot of 35mm and 4x5 sitting in the closet...
Pixel design, CCD or CMOS, and more importantly microlens design can have a measurable impact on the acceptance angle for a pixel and this can impact both vignetting and apparent light loss at very wide apertures. That appears to be what the DXO data is showing.
As a previous poster pointed out I think Mark's idea that the camera manufacturer should not compensate for this is in fact an even less desirable situation. I'd want to be able to assume proper reciprocity behavior in exposure for various aperture and shutter speed combinations. More interesting is that this implies that adapting manual legacy lenses to these sensors will result in a bit of reciprocity failure at wide apertures since the camera is unable to gain compensate without any knowledge of the aperture setting - obviously this would still wouldn't cause an exposure error since most manual users meter stopped down anyway and rarely depend on reciprocity.
In the larger context I believe he's making a mountain out of a mole hill at this point. The 1/3 to 2/3 of a stop is nothing of any concern, and I really haven't met a photographer who buys expensive lenses that are a fraction of a stop faster than another because of the additional speed - rather they are usually going for other qualities of the more expensive lens.
More interesting will be to see if future measurements show any impact on narrow DOF. I suspect that difference will be very small as well, again most people buy the more expensive slightly faster lens because often times the bokeh is just better at all apertures on the more expensive lens.
Anyway, it is an interesting article but I felt it read as if it was written in a rush, not well thought out, and rather half baked at this point. When the focus tests are done and it is properly edited to remove the obvious errors and written more clearly I think it will be a great article. Whether in the end the technical minutia contained will be of any relevance in "real world" photography remains to be seen.
--
Ken W
Rebel XT, XTi, Pany G1, LX3, FZ28, Fuji F30, and a lot of 35mm and 4x5 sitting in the closet...
Cedarhill
Senior Member
If you want to accuse various manufacturers of being dishonest in their representation of camera response to light, then go get a wide sample of cameras of all designs and brands. Shoot all of them at some chosen ISO, aperture and shutter speed. What you will see is at least a stop of difference in the exposure at these known settings. I have noted a 2/3 stop difference in the response of two Nikon DSLRs using the same lens and exact same settings.
Why in the world are you worried about a half stop difference at ISO levels that are seldom used anyway?
Why in the world are you worried about a half stop difference at ISO levels that are seldom used anyway?
Bogdan_M
Senior Member
Thank you for your correction. In fact, CMOS or CCD is irrelevant to the discussion, I should've just stuck with "modern"
You obviously haven't met me and a few of my friends
Seriously, the reason anybody would pay big bucks an a 50 L as opposed to a cheapo 50mm f 1.8 (bout 10X difference in price) is for that extra stop of light and DOF control. The cheapo lens is better by far stopped down. This case is not unique, other focal lengths also have a pair of lenses where the cheaper one is as good stopped down (85mm, 35 mm). My examples are in Canon scope, but I bet the same can be found with Nikon (for 85mm, for instance).
Paying big bucks for 1 extra stop and getting only half a stop + a gain boost to soften my photos even more seems like a deal breaker to me.
Again, it is highly debatable imo. Imagine you have an APS-C system where you are already at a disadvantage in terms of DOF control and therefore you need even more the edge given by a fast lens. What purpose does good bokeh serve if it can't be seen
?Maybe I'm weird, but i had a 35mm f 1.4 for a couple of years, it never heard of f 4, and i have used at f 2.8 just on a few occasions. 90% of the time it was between f 1.4 and f 2.
I'm anxious to see some MFT fast lenses tested. That impressive Voigtlander 25mm f 0.95, I'm curious how much of that f stop really gets some use.
--
http://www.flickr.com/photos/bogdanmoisuc/
Keyboard shortcuts
- f
- Forum