so you are saying
-- it does not matter how much more noise at the pixel level the high resolution sensor has. It must be better as long as you never look at the images actual size - we should disregard the obvious increased noise at actual size
-- no matter how many pixels get crammed on a sensor it is always better even when per pixel noise increases are obvious when viewed at actual size
- no matter how small the resulting pixels get it is always better even when you can see more noise in the image when viewing at actual size
I saw that the 50D was more prone to noise then the 40D - that is especially true at ISO 400 and above - not doubt about it - you can see the difference in images. That fact is confirmed all over the internet. Even Artie at birdsasart was having to do a bunch of extra processing to get the noise out of the 50D images of birds. Hmmm, he must have imagined the extra noise as I did in my images from the 50D.
I have seen is the 5D has much less noise in images at high ISO then the 40D and 50D
that I have seen, so I do believe it
abqnmusa wrote:
so you are saying it does not matter how much more noise at the pixel level the high resolution sensor has. It must be better as long as you never look at the images 100%
I have seen that the 5D has much less noise in images at high ISO then the 40D and 50D
that I have seen, so I do believe it
I think the smaller sensor of the 1.6 camera is the root cause
whatever you want to believe works for me
Think of it this way, as the size of images continues to increase, the relevancy of any one pixel continues to drop. A 100% view is arbitrary anyway, it gradually begins to lose value the larger and larger images sizes become. But again, I'm strictly talking about the IQ of images as a whole, whether viewing digitally or through print. Hopefully this makes sense....
brainiac wrote:
Larger pixels do not give significantly better low light performance. That is a myth. The factors which affect low light performance are sensor size, technological maturity, and manufacturer. This camera would have about the same low light performance whether it was 10 Mpixels, 18 Mpixels or 30 Mpixels. You, and the majority, are confused because you are considering noise per pixel instead of noise per square inch of print. Even famous review web sites like DPR make this mistake, so it is not surprising that many are misled. Noise per square inch of print is the real measure of a camera's ability to capture a picture in low light, and noise per square inch of print is largely unaffected by changing the pixel pitch, whereas it varies dramatically with sensor size, sensor age, and to a lesser extent, manufacturer....Show more →
you are taking the fun out of photography,most if not all DSLRs will print 8x10s or 16x20s with no problems of course you shoot them in RAW and convert to a tiff file...over analyzing pixels is akin to watching paint dry
but most people are satisfied with their 4x6s 5x7s or 6x8s...Analyzing things 99 percent of the buyers have absolutely no clue what the scientists here at FM are talking about...but do they? or should they go out and use the camera for its intended purpose not brickwalls,batteries,charts, ect
sorry sure there may be merit in the over analyzing a pixel to death but I cant
see the value of that...I rather take pictures and sometimes I take pictures with my EOS3..no pixels there
digitalbug30d wrote:
you are taking the fun out of photography,most if not all DSLRs will print 8x10s or 16x20s with no problems of course you shoot them in RAW and convert to a tiff file...over analyzing pixels is akin to watching paint dry
but most people are satisfied with their 4x6s 5x7s or 6x8s...Analyzing things 99 percent of the buyers have absolutely no clue what the scientists here at FM are talking about...but do they? or should they go out and use the camera for its intended purpose not brickwalls,batteries,charts, ect
sorry sure there may be merit in the over analyzing a pixel to death but I cant
see the value of that...I rather take pictures and sometimes I take pictures with my EOS3..no pixels there...Show more →
The problem is that while the average punter correctly believes that more pixels aren't a disadvantage, many camera buyers who express opinions online are labouring under a misapprehension which is encouraged by sites like DXO, DPR et alia, namely that there is a high iso advantage in slowing down the megapixel race. The market is always right, even when it's wrong, so there is a real danger that manufacturers will respond by limiting Mpixels for no good reason except that buyers are victims of a misunderstanding, and incorrectly believe that limiting sharpness offers an advantage which it doesn't. That would be a bad situation for me, because I like really sharp pictures and get paid to take them.
People online need to start realising that there is no conflict between 18 or 24 or 38 Mpixels and great high iso performance. I've tried to show it many times with fair comparisons between D700 and 1Ds3, but with DPR endlessly publishing comparisons of 100% crops from 12 and 21 Mpixel cameras the odds are stacked against the truth emerging any time soon.
brainiac wrote:
People online need to start realising that there is no conflict between 18 or 24 or 38 Mpixels and great high iso performance. I've tried to show it many times with fair comparisons between D700 and 1Ds3, but with DPR endlessly publishing comparisons of 100% crops from 12 and 21 Mpixel cameras the odds are stacked against the truth emerging any time soon.
Have you done this by comparing for example 5D and 5D2 files shot in RAW at high ISO and printed to the same size e.g. A4? With no noise reduction I might add.
As you indicate, dpreview is a classic case of knowing the price of everything and the value of nothing.
And to add to Richard's wise words, if we can't discuss these issues here, then where?
Those "most people" with no interest in the technicalities and the technology are unlikely to linger here anyway (and it's not as if anyone is being forced to read Richard's comments), which leaves the way clear for those of us who really do want to see how things tick.
Coming away from one of Richard's posts having learned something new about the science of the technology doesn't make me any less able to enjoy getting out and shooting my local birds:
There's no "science" in that picture, but it's possible that it is better than it might otherwise have been thanks what I've learned from Richard and folk like him...
brainiac wrote:
The market is always right, even when it's wrong
Wrong from which PoV? Yours? If the market is showing a clear preference to a certain trait/feature then the manufacturers would be silly not to offer products with this trait/feature. It's a very simple business logic. The other option is to educate the market and I'm not sure any manufacturer would want to go into that.
keithreeder wrote:
There's no "science" in that picture, but it's possible that it is better than it might otherwise have been thanks what I've learned from Richard and folk like him...
Nice shot Keith, and thanks for your kind words. I've learned a lot here too, and it's always a pleasure discussing stuff with you and other FM'ers, even when it sometimes gets heated or people disagree strongly.
No - not mine. A market can be factually wrong. For instance, a market might believe that thalidomide is safe. I'm suggesting that the false method of noise comparison on sites like DPR could lead the market to a factually wrong belief. Of course a vendor can go along with a market belief despite knowing that it's factually wrong. For example, a company could go on selling a lethal product even though it knows it's lethal and even though it knows that the market is deluded. Canon may have limited the number of pixels in the G11 sensor because it was the best sensor available from a 3rd party, or because it wanted to extend highlight dynamic range, or because the marketing department has spotted that many G buyers now think, perhaps incorrectly, that fewer pixels will deliver better low-light noise performance. I don't know why the G11 has fewer pixels than its predecessor, but I hope Canon isn't giving up on the megapixel race, because resolution, low noise, and high dynamic range have not proved to be mutually exclusive goals so far.
>The other option is to educate the market and I'm not sure any manufacturer would want to go into that.
Companies spend billions every year educating the market. It's called advertising, and only some of it is pure BS.
Oh it's just "a shot" really Richard - I posted it simply because:
a) it's not unpleasant to look at; and
b) when I was trying to get it, I was drawing on my birding knowledge, fieldcraft skills, awareness of the light and other conditions - nothing (at a conscious level) about the technicalities of how a sensor works and how to manage noise.
But subconsciously I was surely aware of things like the need to expose "correctly" so that I could then manage the dark areas properly and ensure some detail remained without having get bogged down in dealing with noise.
It's information in posts like yours Richard, that has given me an understanding of how to approach shots in less than ideal conditions (it was a dull day and light was at a bit of a premium) and make the most of things.
WillWeb wrote:
I agree with Brainiac for the most part, but to my mind the main advantage of a googlepixel sensor is that you can either print very large and still view from close up, or you can crop significantly. Either way, per-pixel performance has some significance.
I tend to agree here. Given equal print size, the 5DII will win against the 5D hand's down, but what about when we exceed the "maximum" print size of the 5D? After the 5D finds it's print size limit, the higher resolution used to make increasingly larger prints has a per pixel difference.
Let's hear your insight, Brainiac. Looking forward to a change in perspective...
Richard, I agree with you on some points but not on others but I think it can be concluded here. All I can say is that you seem to have nothing to worry about. The trend of increasing MP doesn't seem to slow down.
32067dlm wrote:
I tend to agree here. Given equal print size, the 5DII will win against the 5D hand's down, but what about when we exceed the "maximum" print size of the 5D? After the 5D finds it's print size limit, the higher resolution used to make increasingly larger prints has a per pixel difference.
Let's hear your insight, Brainiac. Looking forward to a change in perspective...
I don't think I can offer any insight as I just don't understand the point. Per pixel performance has significance in precisely the inverse of the ratio of pixels in any two compared cameras. Per pixel noise needs to be 21/12 times better in a 12 megapixel camera in order to match a 21 megapixel camera when 100% crops are compared. But it's almost impossible to judge by the naked eye just when one crop is 21/12 times as good as another. Most people don't understand why they should bias their judgement that way, and they probably can't bias it that accurately anyway. People judge in a gestalt way: this one is better than that one. That's why per pixel judgements are so misleading.
And by the way, there's no print size limit. Print any size you like, the comparison still stands. A small print might hide a better camera's lights under a bushel, but the larger you print, the more apparent the better camera's advantage will be. Going bigger isn't going to somehow reverse the pecking order. It will just consolidate it.
With respect, I think anyone clinging to the notion that per pixel analysis can tell you something that my proposed resized-for-fairness comparison can't tell you, still hasn't got the point.
brainiac wrote:
I don't think I can offer any insight as I just don't understand the point. Per pixel performance has significance in precisely the inverse of the ratio of pixels in any two compared cameras. Per pixel noise needs to be 21/12 times better in a 12 megapixel camera in order to match a 21 megapixel camera when 100% crops are compared. But it's almost impossible to judge by the naked eye just when one crop is 21/12 times as good as another. Most people don't understand why they should bias their judgement that way, and they probably can't bias it that accurately anyway. People judge in a gestalt way: this one is better than that one. That's why per pixel judgements are so misleading.
And by the way, there's no print size limit. Print any size you like, the comparison still stands. A small print might hide a better camera's lights under a bushel, but the larger you print, the more apparent the better camera's advantage will be. Going bigger isn't going to somehow reverse the pecking order. It will just consolidate it.
With respect, I think anyone clinging to the notion that per pixel analysis can tell you something that my proposed resized-for-fairness comparison can't tell you, still hasn't got the point.
For clarity, I don't disagree with you at all, and I realize I can print at 1 pixel per inch if I choose. There's just a lot to get your head around (my head around). You've certainly brought a new paradigm.
I agree with you Richard but there are some minor technical points that you should probably address. First is that a reduced pixel pitch does not necessarily mean smaller sensels. The portion of the imager that we most interested in when we talk about pixel pitch is actually the photo diode and its something whose physical size can remain constant (or at least get only marginally smaller) while pixel pitch decreases. Furthermore, the diode itself can have its sensitivity improved (which Canon claims to have done with the 7D). I realize that these factors are captured under the "technological maturity" factor, just thought I'd put them out there.
I think that comparing a 12mp image to a 24mp at 100% magnification could be considered a fair comparison. If the two images had the same appearance of image quality at 100%, you could say they have the same image quality. You could also say the 24mp is better because there is simply more of the same quality.
Magnification doesn't have to be relative to life size, it could be relative to a pixel, even if the comparing pixels are different size or have different spacing. It doesn't have to be ratio'ed to a dimension, it could be a thing. Just because one sensor packs more pixels into the same size doesn't mean that the magnification reference point has to change. The reference point is chosen by the person performing the comparison. Before people get into discussing which sensors of different types are better than another, they must all agree on the reference.
Uprezzing a 12mp image to 24mp in order to compare it to a 24mp image doesn't sound fair because half the data originates from a processor instead of photons. If your comparison criteria is that the images must be the same mp, an you are allowed to use only one image, then this is all you can do and the criteria considers this a fair comparison.
Things can be looked at a million ways, and the right one is the one that suits your needs.
corndog wrote:
I think that comparing a 12mp image to a 24mp at 100% magnification could be considered a fair comparison. If the two images had the same appearance of image quality at 100%, you could say they have the same image quality. You could also say the 24mp is better because there is simply more of the same quality.
Magnification doesn't have to be relative to life size, it could be relative to a pixel, even if the comparing pixels are different size or have different spacing. It doesn't have to be ratio'ed to a dimension, it could be a thing. Just because one sensor packs more pixels into the same size doesn't mean that the magnification reference point has to change. The reference point is chosen by the person performing the comparison. Before people get into discussing which sensors of different types are better than another, they must all agree on the reference.
Uprezzing a 12mp image to 24mp in order to compare it to a 24mp image doesn't sound fair because half the data originates from a processor instead of photons. If your comparison criteria is that the images must be the same mp, an you are allowed to use only one image, then this is all you can do and the criteria considers this a fair comparison.
Things can be looked at a million ways, and the right one is the one that suits your needs....Show more →
There are right and wrong ways of looking at something. I think your analysis is wrong.
By your unequal magnification method a Minox spy camera can be shewn to produce better quality images than a Hasselblad. The Minox might record a 1cm square image at higher resolution than you will see in 1 square cm of Hasselblad negative. In reality the far greater size of the Hasselblad neg means that it will trash the Minox image every time, even with faster/worse film. If your method of comparison allows the conclusion that a Minox outperforms a Hasselblad, when it obviously doesn't, then there is something seriously wrong with your method. I don't think you have understood this problem.
When you uprez a file the computer is not inventing new data, it is just converting the data that you captured into a finer-grained medium. There is nothing magical about the individual pixels in the 12 Mpixel capture: they have already been interpolated anyway. They are not the sensel values recorded from the sensor. What is more, they are going to be interpolated again before the printer can lay down any ink. Printers don't print pixels, so each pixel has to be interpolated into an ink droplet pattern, i.e. yet another medium. Even on screen you aren't really seeing true pixel values until you view the image at 100%, and even then the pixels in your monitor don't do a very accurate job of displaying the recorded RGB levels. The sanctity of pixels is an illusion. You never really see them. They are designed to be statistical entities, and that is why it is fair to perform the statistical process of interpolation on them. The conversion of their individual values into other values in other pixels is not destructive to the image data itself. Uprezzing does not turn a unicorn into a pantomime dromedary.
A chess board made of 256 squares in colour groups of four still makes a perfectly good chess board. It won't interfere with the game. Similarly, an uprezzed image still shows you all the image the camera captured, it just uses more pixels than absolutely necessary to do so. It's not unfair on the 12 Mpixel file to uprez it because it loses negligible pictorial quality in the uprez process, and the result is the same picture. If you decant a jeroboam into 8 normal decanters you still have all the wine, and it still tastes the same. Uprezzing only changes the vessel/medium, not the message. That's why it's fair.
First off I would have to disagree that referencing a pixel no matter the other bits has to be considered unequal.
I understand your point about the Minox/Hassy reference, and note that I said the 12/24mp comparison could be considered equal if the criteria is to view them at 100% with a pixel as a reference. If the Minox is better at 100%, then it's better at 100%. If the criteria also includes a need for a file larger than 640x427 well then the Hassy wins.
Please recognize that you are considering me wrong simply because of assumed criteria and reference points.
Uprezzing takes data like you say, but in a real photo that data may be different than what the algorithm has calculated. Technically this should result in a file with less detail, though real world use depends on the amount of uprezzing and ones ability to differentiate, and that's why its fairness...depends.
I shoot birds with a 1DmkII and 500/4. Often the birds are too far away and I add a TC and/or crop significantly. One of the factors which limits my cropping capability is the 8 MPx available from this body. When I start shooting with a 7D, I expect to be able (when needed) to crop a lot tighter than I can now, because of the higher pixel density. Since one of my reasons for using the 7D instead of the 1D2 will be to crop more tightly, a comparison based strictly on "resized-for-fairness" would overstate the advantages of the 7D for me - just as a comparison based strictly on per-pixel analysis would understate its advantages.
I'm not saying that more pixels = bad, and I'm certainly not saying that 100% pixel peeping is the proper way to compare an 8 MP body to one with 18 MP. I'm saying that, for me, the correct comparison lies somewhere between that position and yours, Richard.
brainiac wrote:
I don't think I can offer any insight as I just don't understand the point. Per pixel performance has significance in precisely the inverse of the ratio of pixels in any two compared cameras. Per pixel noise needs to be 21/12 times better in a 12 megapixel camera in order to match a 21 megapixel camera when 100% crops are compared. But it's almost impossible to judge by the naked eye just when one crop is 21/12 times as good as another. Most people don't understand why they should bias their judgement that way, and they probably can't bias it that accurately anyway. People judge in a gestalt way: this one is better than that one. That's why per pixel judgements are so misleading.
And by the way, there's no print size limit. Print any size you like, the comparison still stands. A small print might hide a better camera's lights under a bushel, but the larger you print, the more apparent the better camera's advantage will be. Going bigger isn't going to somehow reverse the pecking order. It will just consolidate it.
With respect, I think anyone clinging to the notion that per pixel analysis can tell you something that my proposed resized-for-fairness comparison can't tell you, still hasn't got the point.
