Building up too much expectation is not always good - Canon should know - the 50D seems to be already written off as bad even though the camera hasn't even been released yet. We consumers are rather strange souls!
Yohan Pamudji wrote:
Care to elaborate on your statement about noise? Are you taking final image resizing into account, or are you talking on a per-pixel basis? Judging from your advocacy of comparing files at the same resolution, probably the former?
Noise per pixel isn't the goal. People outside this forum do not buy cameras in order to converse about the merits of 100x100 pixel crops. Nobody should be conned by this per pixel crap:
Salesman: this camera has the cleanest pixels of any DSLR.
Buyer: how many pixels are there?
Salesman: 3 million.
Buyer: goodbye.
Resizing files to the same number of pixels is the only fair way to compare two cameras in 100% crops on your screen.
Koivulehto wrote:
For a company able to make a gapless micro lens array for 38 MP at FF, doing the same for a smaller pixel density, i.e. for larger pixels should be self-evident.
At 20 MP, a FF sensor would have about 1 stop better noise performance than the 1.6 crop sensor of 50D. With 24 MP, it should be about 0.8 stops better than 50D.
For macro shooters any dust reduction mechanism is very welcome - 40D images are almost spotless while 5D images can show hundreds of spots at 3x live size macro shots. A macro shooter doesn't care about AF or weather sealing or FPS when using a flash ... A bright viewfinder would be nice though, but the current 5D one is already adequate for me.
Actually, some info was deleted in an edit. It should have said the pixel density at FF was approximately 38MP. So using the same pixels and performance specs of the 50d, it can't be done.
IOW, we're dealing with a different sensor with different performance specifications. It may be good, but we only have the 1Ds3 to go by on a high res sensor. I'm discounting the Sony completely as I expect Canon to be better, but how much better?
brainiac wrote:
Resizing files to the same number of pixels is the only fair way to compare two cameras in 100% crops on your screen.
Not necessarily true. The real acid test is to...
(a) make prints of equal size
(b) resize files by EXACTLY the same amount, using the same method. When comparing different resolution files this is impossible.
Let me explain (b). Resampling an image is a form of noise reduction. By resizing a 3000x2000 image to 300x200 and comparing against a 600x400 resized to 300x200 is not a fair comparison. The larger image had more 'noise reduction' applied. Which is why (a) is the only meaningful test.
Mark Shaxted wrote:
Not necessarily true. The real acid test is to...
(a) make prints of equal size
(b) resize files by EXACTLY the same amount, using the same method. When comparing different resolution files this is impossible.
Let me explain (b). Resampling an image is a form of noise reduction. By resizing a 3000x2000 image to 300x200 and comparing against a 600x400 resized to 300x200 is not a fair comparison. The larger image had more 'noise reduction' applied. Which is why (a) is the only meaningful test.
Care to explain why A is different from B? Resizing before printing or resiziing before viewing on screen is similar, or?
And if you upres two images to a base standard 50% larger then the largest, you have a quite fair comparison, simple as that (in my book).
Mark Shaxted wrote:
Not necessarily true. The real acid test is to...
(a) make prints of equal size
(b) resize files by EXACTLY the same amount, using the same method. When comparing different resolution files this is impossible.
Let me explain (b). Resampling an image is a form of noise reduction. By resizing a 3000x2000 image to 300x200 and comparing against a 600x400 resized to 300x200 is not a fair comparison. The larger image had more 'noise reduction' applied. Which is why (a) is the only meaningful test.
One could as truthfully say that the lower-resolution sensor made use of "noise reduction" by design, by virtue of its lower resolution. Either way the image with the higher actual resolution will be revealed as such-- in print or on screen.
Well, what do I know. But I'll stick my neck out. Over the next 5 years FF sensor yields are going to go through the roof, to 40 or 50 megapixels.
When that happens, and sensor prices keep on going down, which they will, then one day FF will be as cheap to produce as DX crop is today. At that point DX will cease to have any compelling reason to exist.
The camera manufacturers will then be able to move everyone, serious or not, onto FF. The telephoto/wildlife photographers won't mind theoretically losing the reach of the 1.6x crop multiplier because they will be able to enlarge and crop from the middle of their now incredibly detailed FF images.
The camera manufacturers will then sell lots of FX lenses all over again as DX is quietly withdrawn.
DX was a historical anomaly, an economic, not photographic, necessity.
Well, what do I know. But I'll stick my neck out. Over the next 5 years FF sensor yields are going to go through the roof, to 40 or 50 megapixels.
When that happens, and sensor prices keep on going down, which they will, then one day FF will be as cheap to produce as DX crop is today. At that point DX will cease to have any compelling reason to exist.
Making smaller and smaller elements does not decrease, but rather increases the probably of a defect. The high cost of FF sensors is because so few of them can be physically fit on a wafer, which also raises the cost of throwing one away. You don't get cost benefits of miniaturization with camera sensors because of this.
I don't think that making higher-resolution FF sensors really drags the cost of lower-resolution ones down, either, at least not as much as you may think. Of course prices fall over time as technology is refined, but it's because the manufacturing process is made cheaper, which doesn't really have a direct correlation with decreasing element sizes.
The 40-50 MP number may make more people want a FF sensor-- or maybe not. I have to think that even a stupid consumer will see the light about how much resolution is enough, when they quickly fill up the hard drive of their computer with such large files.
Edited by jvarszegi on Sep 11, 2008 at 07:05 PM GMT
Mark Shaxted wrote:
Not necessarily true. The real acid test is to...
(a) make prints of equal size
Read what I said. The only fair way to compare two cameras ON YOUR SCREEN. Printing is great too, but comparing resized crops is a very very very good guide as to what you are likely to see in a print. Otherwise, how could we work in photoshop? Or do you just shut your eyes and guess?
>(b) resize files by EXACTLY the same amount, using the same method. When comparing different resolution files this is impossible.
Let me explain (b). Resampling an image is a form of noise reduction. By resizing a 3000x2000 image to 300x200 and comparing against a 600x400 resized to 300x200 is not a fair comparison. The larger image had more 'noise reduction' applied. Which is why (a) is the only meaningful test.
Nonsense. Resizing upwards makes very little difference to image detail. Resizing downwards is perfectly fair as long as you remember that the higher resolving file may have been penalised more. I don't know where this "I uprezzed my unicorn image and it turned into a pantomime dromedary" meme came from, but it's basically crap. Otherwise why would we be able to resize up or down before printing without any visible degradation?
jvarszegi wrote:
I don't think that making higher-resolution FF sensors really drags the cost of lower-resolution ones down...
But higher density FF sensors might be cheaper to produce than lower density ones if one can afford to have considerably more dead pixels. For example 4 dead pixels in a 12 megapixel sensor might be more irritating than 16 dead pixels in a 48 megapixel sensor. So the yield for higher densities might increase dramatically.
Edited by brainiac on Sep 11, 2008 at 07:12 PM GMT
I remember how low powered my first computer was in 1995, a 133 Pentium! Nowadays even a cheap computer processor will seriously smoke a high end processor from back them.
DSLR sensors will follow the same rapid increases in power and affordability that AMD and Intel have achieved with computer processors.
The future of DSLR sensors will be comparatively cheap, and extremely powerful, and sooner than we think. Thats why DX is toast, on a 5 year time scale.
...but the miniaturisation benefits of sensors are quite different to those in processors. For a start, bigger sensors are better. Secondly the diffraction limit does mean that pixel shrinking has it's limitations. I don't think we can assume that Moore's law applies in the same way with sensors.
I remember how low powered my first computer was in 1995, a 133 Pentium! Nowadays even a cheap computer processor will seriously smoke a high end processor from back them.
DSLR sensors will follow the same rapid increases in power and affordability that AMD and Intel have achieved with computer processors.
The future of DSLR sensors will be comparatively cheap, and extremely powerful, and sooner than we think. Thats why DX is toast, on a 5 year time scale.
Using the computer analogy, there's actually more to support the argument for crops sticking around for a while.
Yes, sensors have gotten better and cheaper like CPUs have. But to get top of the line computing performance you pay more or less the same amount as you did 10 years ago, albeit top of the line has obviously gotten more powerful, while at the same time there's still a market for lower end processors. I believe the same will hold true for sensors--the top of the line will continue to get better and prices that are currently on the decline will eventually stabilize for that market segment, while there will continue to be a market for lower end stuff that the crop sensors will cater to. FF sensors will always be more expensive than crop sensors, even as manufacturing costs for both continue to decline, so crop will still have its place for the foreseeable future.
I remember how low powered my first computer was in 1995, a 133 Pentium! Nowadays even a cheap computer processor will seriously smoke a high end processor from back them.
DSLR sensors will follow the same rapid increases in power and affordability that AMD and Intel have achieved with computer processors.
The future of DSLR sensors will be comparatively cheap, and extremely powerful, and sooner than we think. Thats why DX is toast, on a 5 year time scale.
They have to be big to be powerful, unlike computer processors. That's a major difference in production cost.
I want the camera you describe too, but I don't expect it to happen, and at the end of the day, it doesn't really matter to my business. I too hate the megapixel wars, but if the 5DII really is 24MP, I'll look forward to sRAW with the knowledge I can go bigger if needed. I'd like more accurate focusing than my 5D delivers, but known advances with the 40D and 50D should provide for a reasonable improvement. I suspect we'll see a 5D replacement with more MP and 50D specs, any more than that is bonus in my book. If I need more, the 1D series is were I'll find it. My glass is half full.
Thanks for your work on that D3 to 1DsIII comparison. I knew that down-sampling reduces noise and comparing images at the same resolution/print size makes the most sense, but for whatever reason I always get caught up with per-pixel noise comparisons. I think the main reason it's still a common yardstick is because it's much easier to quantify with less work than comparing full images with different native resolutions, and for me it's easier to eyeball 100% crops. Also, 100% crops are still valuable for determining the resolving power of a lens-sensor combo for instance, and it's hard to switch gears from using 100% crops for that purpose to using full images for noise comparisons.
brainiac wrote:
But higher density FF sensors might be cheaper to produce than lower density ones if one can afford to have considerably more dead pixels. For example 4 dead pixels in a 12 megapixel sensor might be more irritating than 16 dead pixels in a 48 megapixel sensor. So the yield for higher densities might increase dramatically.
Edited by brainiac on Sep 11, 2008 at 07:12 PM GMT
That's right. But the likelihood of a particular pixel being botched should also rise as pixels get smaller.
We consumers are rather strange souls!