Liquidstone wrote:
I get your point, though IMHO there is neither generally correct nor generally wrong method between "pixel-to-pixel" and "frame-to-frame" comparisons to present noise tests.
For shooters who intend to use the whole frame most of the time (landscapes, etc.), then your "frame-to-frame" comparison is the preferred method.
For shooters (like birders) who often crop their output most of the time, DPR's "pixel-to-pixel" comparo makes sense.
That's exactly right. Or even more specifically, the correct comparison for me is to view at my typical output and print sizes. Everything else is simply gearheaded academics. There's nothing wrong with that, however if the photographer doesn't happen to output at that particular size then the comparison's relevance suffers.
I'm sorry Chris, but Liquidstone's birding comments are precisely wrong. For any given picture with any given lens, a birder will be cropping a fixed proportion of the frame. A sparrow doesn't look bigger with a 600mm lens on a D700 than it does with a 600mm lens on a 5D2. For any given composition with any given lens type, the 5D2 will always put more pixels on the subject, and therefore a comparison of per pixel noise is always proportionately misleading.
The fact that you CAN crop more heavily on a 5D to get more reach doesn't mean that this is how a fair comparison of noise should be done. After all, any camera's noise performance IS WORSE THAN ITS OWN NOISE PERFORMANCE IF YOU CROP MORE HEAVILY. By Liquidstone's and DPR's methods, all digital cameras are both worse and better than themselves. That's how bad the error is.
brainiac wrote:
No, it doesn't, because if you are using a 600mm lens you will be cropping your 5D2 file no more than your D700 file. The 5D2 effective 1200mm crop will still contain many more pixels than the D700 effective 1200mm crop, in exactly the same ratio. This is not about comparing different sizes of sensor, it is about comparing an equivalent area of the frame. Only by equivalent areas of the frame can you compare image noise. That is precisely what DPR and you have failed to do by comparing 100% crops and per pixel noise from cameras with very different pixel densities.
Well, you could be right in your thesis as far as you are concerned.
But personally, I prefer the "pixel-to-pixel" comparison, and I just compensate for differing sensor and pixel sizes depending on the application. As long as the test parameters are well defined and described, I can easily interpret the test results from varying viewpoints.
Use whatever method that works for you, I'm not keen on belaboring this point.
brainiac wrote:
I'm sorry Chris, but Liquidstone's birding comments are precisely wrong. For any given picture with any given lens, a birder will be cropping a fixed proportion of the frame. A sparrow doesn't look bigger with a 600mm lens on a D700 than it does with a 600mm lens on a 5D2. For any given composition with any given lens type, the 5D2 will always put more pixels on the subject, and therefore a comparison of per pixel noise is always proportionately misleading.
Yes, I agree, I think the point I made above was similar. The only truly relevant comparison is to look at identical output sizes (or crops as you refer to them).
Pixel Perfect wrote:
You fail to understand the lighting in the forest - 1/40s @ ISO 3200, it's very dark so of course it looks drab, what were you expecting?
I "fail to understand"?? Really? Is this the way you're seriously going to start this comment? That's pretty lame don't you think? I don't think you really understand my statement. I don't see any color definition, and I don't see any 3D feel to the image. It looks flat. I don't care what kind of lighting situation you are in... my comments are regarding the fact that the noise reduction is robbing the image of these characters that it SHOULD have. There should be more defination to the subject. The noise reduction is robbing the subject of this sort of detail, leaving it looking rather dull.
When I read the title of the post, I was expecting to see an amazing ISO 3200 shot, and I all I see is a ho-hum high ISO shot that looks like any other camera. If you want to see some more amazing high ISO shots, take a look at the D700 or the D3. At 3200, they seem to have more depth despite the heavy noise reduction.
tcamper wrote:
I could care less about what it loks like at 3200. Your example looks great, but this second version better be clean at 12,800, like 3200 is on the original 5D.
Tad
I don't find 3200 clean on the 5D unless I am shooting with a lot of light. I am quite impressed with that bird image. --c
Jeez, you guys go on and on and on and on and on and on and on and on and on and on and on and on and on go on and on and on and on and on and on and on and on and on and on and on and on about this stuff. This isn't fair, that isn't right, that pixel is bigger, that pixel is smaller. It is entertaining and we will all agree, I am sure, that we spend more work time debating and theorizing than we should. I know I am right now
All these newer models produce fantastic results. We are really evolutionary and very incremental now in the improvement category rather than "leaps and bounds" and revolutionary.
One thing is for sure though. When the new 5D2 is released to the general public it will be loved and "film-like" (there is that damn F word again) to half the buyers and "not worth the money" to the other half.
John Power wrote:
Jeez, you guys go on and on and on and on and on and on and on and on and on and on and on and on and on go on and on and on and on and on and on and on and on and on and on and on and on about this stuff. This isn't fair, that isn't right, that pixel is bigger, that pixel is smaller. It is entertaining and we will all agree, I am sure, that we spend more work time debating and theorizing than we should. I know I am right now
All these newer models produce fantastic results. We are really evolutionary and very incremental now in the improvement category rather than "leaps and bounds" and revolutionary.
One thing is for sure though. When the new 5D2 is released to the general public it will be loved and "film-like" (there is that damn F word again) to half the buyers and "not worth the money" to the other half....Show more →
Man, this is worth quoting John... And have been guilty of it too, this gearheaded pixel masturbation that really has very little to do with photography. The "big picture" is that pretty much every DSLR to be released since the 20D has been a remarkable feat of technology and is capable of producing excellent photographs in skilled hands, and we're simply fortunate to be able to use any and all of these tools and toys. And your perspective is very much appreciated, thank you. And for the rest of you that would rather compare pixels and crops than go take photographs and make prints, that's fine too. But I believe this is an entirely (or at least mostly) separate hobby from photography.
ChrisDM wrote:
Man, this is worth quoting John... And have been guilty of it too, this gearheaded pixel masturbation that really has very little to do with photography. The "big picture" is that pretty much every DSLR to be released since the 20D has been a remarkable feat of technology and is capable of producing excellent photographs in skilled hands, and we're simply fortunate to be able to use any and all of these tools and toys. And your perspective is very much appreciated, thank you. And for the rest of you that would rather compare pixels and crops than go take photographs and make prints, that's fine too. But I believe this is an entirely (or at least mostly) separate hobby from photography....Show more →
I took 4300 raw files this weekend. I really don't need to be exhorted to take more. A third of them were taken without flash in a dark night club. That's how I earn my living, so for me it's not my hobby to be interested in every quarter of a stop of high iso performance, it's a matter of professional responsibility. Having taken 4300 pictures, and been paid well to do so, I feel I can afford myself the luxury of enjoying a discussion with the interesting and intelligent folk I meet on this forum, many of whom have taught me a great deal, and to whom I feel a debt of gratitude for their willingness to share their hard-won knowledge and experience. It may not to you, but to some of us every quarter of a stop of high iso performance matters. This is why: http://cyberphotographer.com/1ds3/isomatters.jpg
Many people here shoot landscape, and to them, corners matter. You won't hear me telling them to get over it and enjoy their EF 28's.
>One thing is for sure though. When the new 5D2 is released to the general public it will be loved and "film-like" to half the buyers and "not worth the money" to the other half.
I'm guessing that most of the latter will have compared 100% crops with previous cameras, and most of the former will have compared prints (or crops from files rezzed to equivalent magnification). One group is right, and the other is wrong, regardless of whether they shoot birds.
ChrisDM wrote:
Man, this is worth quoting John... And have been guilty of it too, this gearheaded pixel masturbation that really has very little to do with photography. The "big picture" is that pretty much every DSLR to be released since the 20D has been a remarkable feat of technology and is capable of producing excellent photographs in skilled hands, and we're simply fortunate to be able to use any and all of these tools and toys. And your perspective is very much appreciated, thank you. And for the rest of you that would rather compare pixels and crops than go take photographs and make prints, that's fine too. But I believe this is an entirely (or at least mostly) separate hobby from photography.
Liquidstone wrote:
But personally, I prefer the "pixel-to-pixel" comparison, and I just compensate for differing sensor and pixel sizes depending on the application. As long as the test parameters are well defined and described, I can easily interpret the test results from varying viewpoints.
Others have said they can do this, but I am amazed that you all can accurately judge when one file is 29% noisier than another, and forgive it. I can see when two files are equally noisy, but 29%? Those are some impressive powers of perception you have there. Lucky you.
>Use whatever method that works for you, I'm not keen on belaboring this point.
I am keen on belabouring it, as many people already know, because I think it is very important. Reviews with broken methodology, like DPR's, and the insidious widespread error of comparing 100% crops, are in danger of distorting the market in favour of lesser products, and we, as photographers, will be the losers. That method fails for the great majority of people who aren't capable of allowing for a 29% advantage, even if they are aware that such allowance is necessary.
Not having followed a lot of this, so sorry if this is a really dumb question, if you are saying that you must resize the files to the same dimensions to evaluate noise properly, does it matter if you uprez or should you always downrez the higher MP file?
brainiac wrote:
I took 4300 raw files this weekend. I really don't need to be exhorted to take more. A third of them were taken without flash in a dark night club. That's how I earn my living, so for me it's not my hobby to be interested in every quarter of a stop of high iso performance, it's a matter of professional responsibility. Having taken 4300 pictures, and been paid well to do so, I feel I can afford myself the luxury of enjoying a discussion with the interesting and intelligent folk I meet on this forum, many of whom have taught me a great deal, and to whom I feel a debt of gratitude for their willingness to share their hard-won knowledge and experience. It may not to you, but to some of us every quarter of a stop of high iso performance matters. This is why: http://cyberphotographer.com/1ds3/isomatters.jpg
Oh don't get me wrong Richard, high ISO performance is very important to me too, that's why I shoot all my low light/event work with a 1D3. Its ability to shoot at 6400 helps to mitigate motion blur in situations like the one you posted above. It would have at least doubled your shutter speed.
if noise at high ISO is all you care, 5D with noise ninja/neat image should produce less noise/on par with 5D II with only $100 upgrade (but u dunt get all the excitements with new toy which 75% of reason ppl make the upgrade )
Care to show me how to switch a 5D's iso to 6400 and 12,800? Please don't respond by telling me to push it in PP.
scott f wrote:
If you are saying that you must resize the files to the same dimensions to evaluate noise properly, does it matter if you uprez or should you always downrez the higher MP file?
This thread says it all (see p. 1, #8 re: your question, Scott), and it should be required reading by anyone (including review sites like dpreview) who wants to do comparisons between different-megapixel cameras wrt high-ISO noise:
Care to show me how to switch a 5D's iso to 6400 and 12,800? Please don't respond by telling me to push it in PP.
Tad
I won't tell you to do it, but that's how the cameras do it already. 3200 on the 5D is 1600 pushed a stop in-camera, and 6400/12.8k are pushed 3200 by 1 and 2 stops respectively in-cameraon the 5dm2.
Guys, Richard is definitely right about the noise comparison methods. Since we are comparing 2 different 36x24mm sensors taken with ostensibly the same focal length lens, the size of the pixels is misleading for noise or even image quality comparisons.
Let me try to explain the fallacy from another angle, and forget that we are talking about noise and just talk about image detail. When you make an 8x12 print at 240dpi and compare that to a same-sized print of the same image downsized to 120dpi. The lower dpi image will obviously have less detail when viewed at the same distance. Put another way, the detail goes level goes down because the size of the individual "dots" goes up. Now take that downsized image and make a 4x6 print, viewing the 4x6 print from a closer distance so that the apparent size of the 4x6 print and the 8x12@240dpi print is the same. Suddenly the 2 images seem to have the same level of detail, which of course is not really true. In the end customers want a print of a given size, not a print with a given DPI they are going to view from a particular distance.
Now using that same thought experiment, let's talk about image noise from two different images with exactly the same framing taken with exactly the same lens and exposure settings, but taken with 2 differnet density sensors. If we say that the 2 cameras (5d and 5dm2) have the same per-pixel noise, to me that means over a given 100x100 pixel portion of a scene the number of noisy pixels is the same. Since the 5dm2 has a higher pixel pitch, the relative size of each noisy pixel is going to be smaller than a 5d image. Put another way, if I make an arbitrarily sized print from both images, the size of the noisy pixel will be larger on the 5D print. The larger the noisy pixel, the more obvious it becomes.
Pixel Perfect wrote:
Why would you choose the 5D II for FL limited birding anyway. Surely you'd have a 50D or at least 40D in the bag. You can't knock the 5D II if your cropping away 13MP. Try the same crop on a 5D at high ISO and see how it compares.
Why would I bird with a 5D II? Simple.. it makes life a lot easier...
Since it has the same pixel density as the 20D, so I won't be losing any 'reach', so that's out of the equation. Why would I not use a 40D or 50D instead? Because being able to compose with my bird center frame so I can just use the center AF point on a FF 5D II makes focusing a lot easier than if I had to focus/recompose as I would with a 1.6x sensor. With the 5D II when I crop I'll still have plenty (ie. 8) MP's left. Also I haven't been terribly impressed by the 40D or 50D, and rather save the money for owning a 1 series in the future.
The answer to this is not having more AF points, because I can't change AF points as quick as a bird can turn it's head, and turning the head will mean I want to put it on the opposite side of the frame so it is looking 'into' the picture. Also... small birds move quick, and you never know what side of the frame it's going to be on 1 second from now.. Also with BIF and changing backgrounds it's a lot easier to just keep using the same AF point. Not to mention the center AF point is always better. I'd love to have a 1DsmkIII for this, but it's a little beyond my price range, so I'm excited about the 5D II being affordable for me. If the AF is as good as the 20D, it's plenty good for me, in low light sometimes I wish it was a little better, but it's certainly no deal breaker.
Note that I don't do a lot of action stuff, so frame rate isn't critical. For the action stuff I do do, timing trumps fps. I'd of course love to have a 1DIII for when I want to do action work... but I can't at the moment afford to get both that and a 5D II.
So, when it comes down to it, while the IQ of my images won't change much compared to what I'm getting out of my 20D, BUT using a FF camera with the same pixel density will make my life a lot easier when it comes to focus and composition for quick moving birds. My only disappointment is that I would have thought Canon would have figured out how to increase the sensitivity of their individual pixels by more than 1 stop.. but as Liquidstone said, perhaps we're reaching the point where for the architecture and materials currently available we've simply hit a wall in sensitivity.
brainiac wrote:
I am keen on belabouring it, as many people already know, because I think it is very important. Reviews with broken methodology, like DPR's, and the insidious widespread error of comparing 100% crops, are in danger of distorting the market in favour of lesser products, and we, as photographers, will be the losers. That method fails for the great majority of people who aren't capable of allowing for a 29% advantage, even if they are aware that such allowance is necessary.
Belabor on, then.... you won't hear any argument from me.
Once your method gets widespread acceptance, then I'd switch to your way of doing things. English might not be the best language ever invented by man, but I have to use it because many more people on the forums I frequent understand it better than say Ilocano. The "pixel-to-pixel" way of comparing things might not be perfect, but it has been a frame of reference for some time, and if you google "50d noise test," most reviews that come out used 100% crops, including the links below:
)