Tentacle wrote:
The DPReview DR test on the 5D was flawed because it used JPG shots. The JPG still had recoverable image data in it which suppressed the measured DR. Other 5D tests have shown it has a better DR than reported in that DPR review, between 9 an 10 stops.
I was actually referring to Phil's "RAW Headroom" testing which used a RAW image, but is subject to some subjective interpretation. His "ACR best" test showed the 40D at 10 2/3 stops and the 5D at 10 stops, which would lead me to believe that the 40D has a 2/3 stop advantage (assuming his gray scale is devided into 1/3 stop increments), but in practice, the 5D has the latitude.
The problem with the dark side of the DR range is image quality. If you accept 75% accuracy you need to be just 4 times above noise floor. This is half a stop in deviation. Narrow it down to accepting only 0.1 f-stop in noise, which means better image quality, and DR shrinks. The quality of the RAW converter also makes a difference.
Phill would have to re-do the 5D DR tests in order to compare it to the 40D properly.
I will study your links a bit more when I get time.
I think Phil's test isn't an accurate depiction of useable DR either, though I commend his efforts and his willingness to show his work. I agree that he doesn't give a lot of respect to the noise floor in his measurement (and I wish he'd revamp his noise comparisons as well to show a standard image with greater micro-detail than the one he uses now).
I'd love to see testers settle on a standardized DR test (perhaps like Imatest) that would produce a repeatable number that could be used to compare across brands so that the buyer could actually see what the useable DR of a given camera is, taking into account noise and highlight retention.
Tentacle wrote:
I'll try to gracefully bail out of this part of the discussion. Dynamic range is the difference between the smallest accurately measurable signal and the highest measurable signal.
It's, in this case, the difference between the lowest measurable signal above noise floor and the signal at saturation. These numbers are fixed/constant. If you view a whole image or
just 10x10 pixels it makes no difference. Each pixel will still saturate after the same amount of photons. It doesn't matter if I look at 1, 100 or 1600x1200 of them.
If you can't grasp this concept about DR, then this particular discussion is aborted....Show more →
Let me see if I can put it in the terms that you like. Take one big pixel, which saturates at 60000 electrons. Now consider four pixels filling the same area, doubling the pixel pitch. Each one need only saturate at 15000 electrons for the combined pixels to have the same capacity as the big pixel. The sources of read noise also scale approximately with pixel pitch (as the 1D3/40D comparison illustrates) and so the read noise per pixel of the small pixels will be about half that of the big pixel; for instance if the read noise of the big pixel were 20 electrons, the read noise of the small pixels would be 10 electrons.
So you are correct, the small pixels have a DR of 15000/10=1500 in this hypothetical example, while the big pixel has a DR of 60000/20=3000, twice as much. But suppose I do the simplest, coarsest processing and simply bin together four of the small pixels. Combined, they hold 4x15000= 60000 electrons total, and since noises combine as the square root of the sum of squares, the read noises aggregate as sqrt[4*10^2]=20, the same as the big pixel. The image with small pixels binned has binned pixels with the same resolution and the same DR as the original big pixel camera.
Just because smaller pixels individually have less DR does not mean that when compared on the same scale as the big pixel they have less DR in aggregate. In fact for the 40D/1D3 comparison the DR's are equal when compared fairly, as I showed from measured data.
People who moan about the megapixel race ignore this fact, look at all images at 100% magnification, conclude that the individual pixels have more noise and less DR, and moan some more. They ignore the fact that when the image is resampled to print dimensions, pixels are combined in a way that levels the playing field as far as noise and DR at low ISO, with the added benefit of more resolution for the smaller pixel camera.
jamesf99 wrote:
I guess I disagree though as I would much prefer a native ISO 50 option, let alone ISO 100. I've been in situation where I was fully stopped down, max shutter speed, and didn't have a ND filter with me. Even 1 more stop would have helped, but ISO 200? Nah.... I want some "slow time".
Different strokes for different folks. I would only say that there is an aftermarket add-on in the ND filter that can lower your effective ISO, but there is no aftermarket add-on that can give you clean ISO 3200 starting from an inefficient sensor with correspondingly low base ISO.
ejmartin wrote:
Different strokes for different folks. I would only say that there is an aftermarket add-on in the ND filter that can lower your effective ISO, but there is no aftermarket add-on that can give you clean ISO 3200 starting from an inefficient sensor with correspondingly low base ISO.
I'm not disagreeing with what you said, I simply would prefer it starting lower as opposed to higher. There is no substitute for the quality of slow film, and the analogue appears true with digital, but I didn't get into your discussion above
One thing that has always struck me as funny is that Canon seems to under report their ISO ratings. IOW, when Canon says ISO 100, it's really ISO 150-175, and so on up the later (if IIRC, the error compounds with each jump). When you compare Canon's IQ at a true ISO, they are even more favorable than they appear.
Hope that the new 29 points AF will have larger frame coverage. Stay tuned until at least this Summer guys. Canon must wait the EU promotion to finish.
Good point - our rebates don't end till 30th June. Although, I suppose you could argue that if the theoretical new camera was released before then, it might be a incentive to get a 5D while you still can - and at a damn good price. Perhaps it's a good potential move on Canon's part to clear old stock. Or maybe it's just BS
Funnily enough I recieved my cheque from Canon today for £100 against the 70-200/4 IS. The current rebate is for only £45. Tight buggers... I was thinking about getting a EF-S 60 macro - but it's not included!
Let me see if I can put it in the terms that you like. Take one big pixel, which saturates at 60000 electrons. Now consider four pixels filling the same area, doubling the pixel pitch. Each one need only saturate at 15000 electrons for the combined pixels to have the same capacity as the big pixel. The sources of read noise also scale approximately with pixel pitch (as the 1D3/40D comparison illustrates) and so the read noise per pixel of the small pixels will be about half that of the big pixel; for instance if the read noise of the big pixel were 20 electrons, the read noise of the small pixels would be 10 electrons....Show more →
Read noise for current CMOS sensors is about 4 or 5 electrons per pixel. Regardless of pixel size, as clearly shows in the graph. Data in the table shows that the 1D3 has 4 read noise electrons, as opposed to 4.2 for the 40D.
So you are correct, the small pixels have a DR of 15000/10=1500 in this hypothetical example, while the big pixel has a DR of 60000/20=3000, twice as much. But suppose I do the simplest, coarsest processing and simply bin together four of the small pixels. Combined, they hold 4x15000= 60000 electrons total, and since noises combine as the square root of the sum of squares, the read noises aggregate as sqrt[4*10^2]=20, the same as the big pixel. The image with small pixels binned has binned pixels with the same resolution and the same DR as the original big pixel camera.
Once you start binning you average values. That means that the lowest value rises and the highest value falls. Result? Decrease in DR.
Just because smaller pixels individually have less DR does not mean that when compared on the same scale as the big pixel they have less DR in aggregate. In fact for the 40D/1D3 comparison the DR's are equal when compared fairly, as I showed from measured data.
Again a conflicting source, same link as above. 40D and 1D3 DR are not the same.
People who moan about the megapixel race ignore this fact, look at all images at 100% magnification, conclude that the individual pixels have more noise and less DR, and moan some more. They ignore the fact that when the image is resampled to print dimensions, pixels are combined in a way that levels the playing field as far as noise and DR at low ISO, with the added benefit of more resolution for the smaller pixel camera.
Resampling only goes one way: DR can only stay the same or it can go down. DR can not increase as you resample. A low res image from a high DR camera will take a small hit in DR, but it will retain it's range. A high res image with low DR can never restore high DR, that's image data that wasn't recorded in the first place.
Mar 11, 2008 at 03:18 PM
Steve Spencer Online • • • • • • • Upload & Sell: On
pumbaa wrote:
Hope that the new 29 points AF will have larger frame coverage. Stay tuned until at least this Summer guys. Canon must wait the EU promotion to finish.
I doubt the will have much more coverage. Chuck Westfall's tech tips from today http://digitaljournalist.org/issue0803/tech-tips.html suggest it really isn't possible to get much more coverage using Canon's approach to autofocus.
jamesf99 wrote:
One thing that has always struck me as funny is that Canon seems to under report their ISO ratings. IOW, when Canon says ISO 100, it's really ISO 150-175, and so on up the later (if IIRC, the error compounds with each jump). When you compare Canon's IQ at a true ISO, they are even more favorable than they appear.
You are correct up until the newest generation of cameras (400D/40D/1D3/1Ds3), when Canon finally decided to get with the program and adhere to the industry standard for ISO. Before that, Canon stated ISO was 1/3 stop faster than the standard, ie stated ISO 100 was actually ISO 125, stated ISO 200 was actually ISO 250, etc. Hence as you say, the IQ was understated as well. But with the latest generation the ISO calibration is accurate, and the advantage has disappeared (making at the same time the current generation appear better performers relative to their predecessors by 1/3 stop even if that was just relabelling the stated ISO).
Read noise is 4-5 electrons at ISO 1600. At ISO 200, which I was discussing, it is what I quoted -- 12.3 electrons for the 1D3, 9.6 for the 40D (at least on the samples I tested; there can be about a few percent sample variation). Clark's data on the two at ISO 100 show 24.4 electrons for the 1D3, 20.1 for the 40D. Within a few percent, low ISO read noise appears to scale as the pixel pitch. And that's where you should consider it, since the max DR occurs at low ISO.
As I mentioned in previous posts, the fact that high ISO read noise saturates at 4-5 electrons in current Canons means that small pixel cameras suffer from a higher noise floor at high ISO; but they do not suffer that at low ISO, rather they have a lower read noise commensurate with the pixel pitch.
Once you start binning you average values. That means that the lowest value rises and the highest value falls. Result? Decrease in DR.
Um, no -- the result you just described is that the *fluctuations* around the mean have decreased via binning, in that binned values move closer to the average value; ie the noise has decreased. And thus the S/N ratio has increased. Precisely my point.
Again a conflicting source, same link as above. 40D and 1D3 DR are not the same.
Clark reports pixel level DR. I readily concede that smaller pixels have lower pixel level DR, as my hypothetical example shows, and as is true of the 1D3/40D example. DR at the same spatial scale can remain constant, as the hypothetical example shows and the 1D3/40D real world example demonstrates unequivocally.
It never ceases to amaze me how some of these threads turn into techno babble...
Donning asbestos suit prior to hitting the send reply button.
As to the original topic of this thread, I would love a 5D or it's successor and will one day purchase one, I (as are many others) looking for a small form factor 1 series camera, I am heavily invested in Canon glass and don't plan to change that. I am still making nice photographs with my lowly 20D.
Clark reports pixel level DR. I readily concede that smaller pixels have lower pixel level DR, as my hypothetical example shows, and as is true of the 1D3/40D example. DR at the same spatial scale can remain constant, as the hypothetical example shows and the 1D3/40D real world example demonstrates unequivocally.
We'll leave it at that, shall we? This whole concept of spatial DR feels bogus to me, because what it does (or rather, the way I understand it) is that resampling will lower the effective noise floor, thus allowing a little expansion at the dark end of the dynamic range. That feels like an artificial procedure, a trick, it doesn't change the pixel DR. But now we're sorta into semantics.
It doesn't change the fact that on the bright side of the DR a bigger photon well will allow for more light before saturation happens. Since noise isn't an issue on that side of the DR, resampling will not help.
For the benefit of those who may have been missed this little bit of info about Canon's CMOS full frame sensors. It is not new but I am sure lots of folks might have overlooked it.
i will admit that those specs looks fantastic and actually look quite similar to what we have been saying that the new 3D should be like. regardless of the specs and how much i would love to buy this camera, i will not touch it with a 10 foot poll when it comes out anyway after the mk3 fiasco
Further to that thread at dpreview, it is said Canon will announce a new flagship at photokina but it won't go on sale until PMA 2009. Taking a leaf out of Nikon's book, and announcing a product months before it's ready. Probably a preemptive strike. Make the specs so impressive it'll take the wind out of Nikon's D3X sails and get people waiting to see how it performs.
