Spin off thread to keep Alex's Canon lens thread on topic.

Subtitle: Every man has a price and a breaking point.

I’m a bit afraid if loyal Nikon shooters switch so easily.

I don't want to drag Alex's thread off topic, so I'll come over here...

I will say that (for this Nikon shooter at least, and I'm pretty sure for others) it is NOT an easy switch, certainly not something considered lightly.

I've been arguing backwards and forwards (both with myself and with James) re: possible switching for at least 10 months now. I have put off lens purchases that I had planned because I was thinking of jumping ship, I haven't bought an SB800 to go with the D70... It just goes on and on. And I still don't know what I am going to do.

Everybody has their crunch point I suppose. For me, I'm planning a landscape shoot in the far north of Australia, followed (maybe) by a couple of days in Tulamben, in 7 months time. I want (note, I don't say need, I am sure I could get by with my current D70 and another D70 body as a backup) a pro/semipro, higher res body with a bigger viewfinder and better AF than the D70 for that trip. For me, crunch time is rapidly approaching.

Well, I've gone through the same mental gymnastics. I've wanted another digital body. I no longer carry the F100 as backup. Carrying film I never use is a pain. I find the D70 lacking in certain features. But I have to ask:

Switch to what?

If it is the 1DsMKII, then OK, there is no equivalent.

If it is the 1DMKII, then why not wait for the D2X?

Since I really have no intention of spending $8k (approximately 68,000,000 rupiahs), the 1DmkII would have been my choice, but the D2X may be a better choice. It won't match the 1D for speed and likely noise performance, but 12mp is darn good as will be the CAM2000 AF and the 10000000 point matrix metering. I-TTL is supposed to be excellent. The 1.5x crop factor is OK by me, since the wide end is covered by the 12-24mm.

I'll just sit tight for now.

Everybody has their crunch point I suppose. For me, I'm planning a landscape shoot in the far north of Australia, followed (maybe) by a couple of days in Tulamben, in 7 months time.

Well Craig, since a large reason for me buying gear is for landscape shoots in the far north of Australia, perhaps my input will be of value.

Dare I say it, Nikon DSLR gear in early 2005 is not a good choice for landscape shooting. 1ds, 1ds Mk2, and P25 are the cameras you really want for this. 1Ds is not really quite good enough, IMHO. 1Ds Mk2 is hampered by poor wide angle lenses. p25 is just too expensive, unless you happen to be a vetinary gastrroenterologist in which case you will probably buy 2 or 3.

The call is whether to change systems, or shoot medium format film, or try one of those crazy Kodak DSLRs. Decisions, decisions. And it depends on your output - my above comment really applies only if you want the best possible quality, which traditionally has been the case for landscape.

For underwater, I still think Nikon is a viable choice and I'm actually a bit suprised at the sudden migration of Wetpixel Nikonians over to Canon of late.

If you're shoot is taking you to the northwest, send me a pm and we'll go for a dive.

Scorpio: Interesting interpretation. If the D2x and 1DmkII are about the same, that for me reinforces the switch to Canon. It's because I can use the same housing when I someday upgragde to a 1DsmkII. Will the D3 body fit in the D2x housing? Who knows, right?

The camera upgrade path for the 1D series housings is very good, IMO.

Cheers

James

One of my main reasons for considering Canon is that they seem a better company to me. They are leading both in terms of high end technology. And (I guess because they are bigger) Canon unlike Nikon they have not had the supply/slipping deadlines issues that Nikon have had with their last two cameras (D70 and D2X). I'd rather be allied to the more efficient company.

My interest in Canon is in the 1DS MkII more than the 1D Mk2. Afterall I have two kidneys.

I suspect that when it appears we will find the D2X is pretty similar to the 1D MkII. The 1D MK2 leading at high iso performance (probably by quite a bit) and the Nikon leading in resolution and wide angle optics. There is not a full frame fisheye for the 1.3 Canon crop is there? For my type of UW shooting I value a 180 degree fisheye and (less significantly) resolution above high ISO noise performance (I took 25000 images UW with my D100 and 0 of them was taken at above the lowest ISO).

I think it is a sigificant point that the 1D MK2 offers a camera upgrade in the same housing option. Especially as this gives you a spare body. But I also don't see Canon using that body again now. Most people place the 1D series ahead of anything offered by Nikon in all areas apart from the body.

After reading Thom's page (following James's advice) I note that Nikon say they could already build a 20MP FF digital. But they think it is too expensive for the market right now. But if this camera does become the D3 (in 2-3 years) I think that there is more chance it will be based on the D2 body than the F6 body, and therefore might just fit in a D2 housing. And while I am dreaming maybe it will come with a free Porsche.

I can't believe how much speculating I have done above! Maybe I should shut up and wait to see how good the D2X is!

Alex

These questions are naive; I am still futzing with a CP5000 and still only planning, reading, saving, and dreaming about the DSLRs. But, what do the cognescenti think about...

1. Nikon seems very focused on the 1.5 crop. This is bad for FF folks, but good if the 1.5 size is okay for your application. On the other hand, Canon is diluted across 1, 1.3, and 1.6. The bottom line: the DX lens line looks much more forward-compatible than EF-S, doesn't it? And isn't this useful, especially with the danged good, reasonably priced Nikon 10.5 FE,

the impressive 12-24, and their other lenses (plus compatability with Nikon's very good non-DX macro lenses)?

2. How far are we from getting to (almost universally) acceptable noise levels and dynamic ranges on 1APS sensors? One year? Three? I doubt it's more than 3. It seems Canon is VERY close (with the 20D), and it's not unreasonable to believe that Nikon will get closer soon. And the resolutions are pretty well there. So, especially given the potential for cheaper high-quality optics and adequate WA with APSish sensors, does it really make sense for most of us (the truly accomplished and wealthy excluded) to switch camera lines for a larger sensor? Or to abandon thousands in lenses to get a quieter sensor this year, when we could have kept our lenses and waited a year to get the same performance? Or, put another way, will FF shooters have a discernable advantage over APS shooters 3 years from now? Or, yet another way: Will APS shooters be able to draw on a wide range of DX or similar lenses, leverage more broadly sold sensor technology, and therefore be able to have perfectly acceptable sensors and lenses, and more compact equipment, at a lower price (allowing purchase of "this year's technology" more often)?

3. How far are we from getting to USB (or similar, flexible interface) driven camera controls? These are needed for some lab and studio apps, and Nikon Capture (and I imagine Canon, too) already supports a lot of remote control. Couldn't housing manufacturers use these capabilities to build much more flexible, almost body-config-independent housings to mitigate the upgrade problem (similar to what L&M did with the D100, but reflecting each manufacturer's aesthetics for look and feel)?

I know for the truly dedicated shooters, 3 years is too long to wait -- but that's how slow MY upgrade cycle is.

It's because I can use the same housing when I someday upgragde to a 1DsmkII.

By the time you upgrade, it will be the 22mp 1DsMkIII and you can't be sure it will fit. :wink:

Will the D3 body fit in the D2x housing?

No worries, the D3 is at least 3 years away from being anounced and 5 years from being available.

I suspect that when it appears we will find the D2X is pretty similar to the 1D MkII. The 1D MK2 leading at high iso performance (probably by quite a bit) and the Nikon leading in resolution and wide angle optics. There is not a full frame fisheye for the 1.3 Canon crop is there? For my type of UW shooting I value a 180 degree fisheye and (less significantly) resolution above high ISO noise performance (I took 25000 images UW with my D100 and 0 of them was taken at above the lowest ISO).

Your opinion that high ISO performance is not important because you never shoot there is a common one, and one that I disagree with. I've been meaning to take a little time to explain my view on this, and this seem to be a good time. Here's the familiar noise chart from dpreview:

Each of the data point on this chart comes from an exposure in the middle of the exposure range. If he did a good job on the test, each successive point on a curve came from a region that got half the light of the one to it's left. Another way to interpret this is to think of this as coming from one exposure and the point at twice the ISO as the lowest one is just one measured at a place in the image that got half the light as the one from the mid point and so on. Thinking about it this way, you can expect the noise on a IDMkII image from a region that is 4 stops darker than the mid exposure point is about the same as that from the mid exposure point from a D2H image. Although for this to be true the IDMKII has to be shot at 2 stops higher exposure. We're assuming getting enough light is not an issue.

Various noise reduction algorithms built into the camera will affect my analysis above a little, but I don't think it would change the basics much.

Herb, what about programmable ISO gain? How a camera performs at ISO 800 has no bearing on how it does at ISO 100. They are not being operated the same way. I think you're mistaken in your conclusions.

Phil's charts are pretty and they're useful for comparing cameras on the chart, but the charts are unitless and the results for a camera vary from review to review. You have to wonder how exactly Phil arrives at these numbers. Where are you getting that "Each of the data point on this chart comes from an exposure in the middle of the exposure range"? I don't understand that statement.

I think for most cameras whose noise doesn't vary with exposure (expose right applies) it's safe to say that high ISO performance is irrelevant when you're shooting at base ISO. The analog noise reduction circuitry is not operating the same way in those two cases.

Another way to interpret this is to think of this as coming from one exposure and the point at twice the ISO as the lowest one is just one measured at a place in the image that got half the light as the one from the mid point and so on. Thinking about it this way, you can expect the noise on a IDMkII image from a region that is 4 stops darker than the mid exposure point is about the same as that from the mid exposure point from a D2H image.  

I'm sorry Herb, but that just strikes me as totally wrong. And here is my possibly skewed logic:

The noise measurements probably are based on a mid-grey/18% grey value (assuming Phil has a reasonably sensible analytical method), but a tonality that is one stop lower than mid grey, and thus is receiving half the light input, will render as one stop lower than mid grey, and after 3 stops or so down from mid-grey, you'll be at black. As you should be. In order for your argument to make sense, the sensitivities would have to be altering on a pixel by pixel basis. You imply that the shadows in an image are amplified differently than the highlights, and thus that a 2 stop below midtone image has the noise of a 2 stop higher ISO rating.

Increasing the ISO rating by one stop (say 100 to 200) increases the amplification of all pixels evenly, and then the mid-grey value is rendered as a mid-grey, as it should be, but in response to half the number of photons. One stop down from mid-grey in this scenario is 1/4 the number of photons of one stop below mid-grey at ISO 100, and so the noise (dark current, thermal heating or whatever, but it is constant for a constant shutter speed) is increased.

And as I was typing this, I see the other Craig is leaping into the fray.

Herb, what about programmable ISO gain?  How a camera performs at ISO 800 has no bearing on how it does at ISO 100.  They are not being operated the same way.  I think you're mistaken in your conclusions.

By programmable do you mean where the camera picks an ISO or just setting that they are quantized in smaller units than one stop? In either case, it is just setting an amplifier gain. I think I made this argument with you before. The signal-to-noise at ISO 800 at any given brigtness in the image should be less than or equal to signal-to-noise at a brightness that is 3 stop lower at ISO 100. Otherwise just shoot at ISO 100 and bump it up 3 stops in Photoshop.

Phil's charts are pretty and they're useful for comparing cameras on the chart, but the charts are unitless and the results for a camera vary from review to review. You have to wonder how exactly Phil arrives at these numbers. Where are you getting that "Each of the data point on this chart comes from an exposure in the middle of the exposure range"? I don't understand that statement.

Phil explains his method at the top of the page with the graph I showed. This is from the 20D review and is changed somewhat from the 1DMkII review, so you are right they cannot be directly compared. I was only making comparisons of cameras on that chart. I think his method is ok, given that he's shooting jpeg with usually the default settings. At least I think they give a good rough estimate of sensor performance if you interpret it carefully.

To measure noise levels we take a sequence of images of a GretagMacBeth ColorChecker chart (controlled artificial daylight lighting). The exposure is matched to the ISO (ie. ISO 200, 1/200 sec for consistency of exposure between cameras).

 

 

The image sequence is run through our own proprietary measurement tool which measures the standard deviation (normalized) of the middle gray patch (indicated by the red rectangle above). Additionally we now have a 'detail crop', this is currently a postage stamp (lots of fine detail) but we may replace this at a future date. Note that noise values indicated on the graphs below should not be compared to those in other reviews.

By "middle of the exposure range" I meant the patch at mid grey, which probably is not the best way to say it.

I think for most cameras whose noise doesn't vary with exposure (expose right applies) it's safe to say that high ISO performance is irrelevant when you're shooting at base ISO. The analog noise reduction circuitry is not operating the same way in those two cases.

To address this, I'll have to go deeper into the "technical mumbo jumbo" as one of our friends put it. Setting the problem of dark current aside, ther are two main contributions to the noise. One part is due to amplifer and for CMOS also device mismatch (pattern noise) that is not dependent on how many photon the pixel sees. The other part is just the statistic of the light itself, the shot noise, which is just equal to the square root of the number of electrons in the bucket at the end of the exposure. You can get the numbers for some of these chips from the spec sheet. Kodak is very good at spelling out the details on some of their stuff. The numbers I recall off the top of my head for the sensor that went into the E-1 is RMS noise leve l ( the fixed part) of about 10 electrons, and a well capacity of 8000 electrons. That gives a mid exposure of 4000 electrons and shot noise of 63 electrons at that exposure. At the lowest ISO at mid grey, the exposure dependent part dominates, the fixed part becomes a bigger fraction at higher ISO and darker places.

The poorer performers have either a small well or larger background noise level or both, and it can shows up in a single image at the lowest ISO if that image has large dynamic range.

The noise measurements probably are based on a mid-grey/18% grey value (assuming Phil has a reasonably sensible analytical method), but a tonality that is one stop lower than mid grey, and thus is receiving half the light input, will render as one stop lower than mid grey, and after 3 stops or so down from mid-grey, you'll be at black. As you should be.  In order for your argument to make sense, the sensitivities would have to be altering on a pixel by pixel basis.  You imply that the shadows in an image are amplified differently than the highlights, and thus that a 2 stop below midtone image has the noise of a 2 stop higher ISO rating.

You are right that 3 stop down from mid grey is close to black the way the camera usually renders the image. But we shoot raw and apply our own curves to most of our shots. Here's where the information captured 3 stops down from mid grey may be important: If you're shooting into the sun and have to radically decrease exposure to get the sunball to come out. Some of the blue water may end up at a brightness 3 stops down from mid grey, and can be brought out without clipping any more of the sunball by using curves. So it doesn't have to be pixel by pixel, just has to be non-linear.

I'd agree with 2Oceans that of course underwater photography technique is far more important than worrying about cameras! I guess it just crops up for me when - like now - I am between cameras.

Interesting discussion between Herb and the Craigs. I kind of end up either a bit lost or agreeing with both sides!

Alex

I'm kind of hung up on the "our own proprietary measurement tool" part! I think Phil does a good job on his tests and I trust him, but the noise graphs cannot be compared from review to review or compared to data from any other source.

I think you said it right, Herb. There's a reason why selectable ISO is offered (I said programmable before) and if it's not to minimize the impact of noise I can't imagine what it would be. We should see some straight lines on the ISO noise graphs if there wasn't some significant noise reduction at higher ISO's going on. I just think your conclusions backwards from that. How can a result from a high ISO measurement tell you anything about base ISO performance when the signal processing is not the same?

Hi guys, very interesting discussion! I agree with both of you on this one - Phil's charts are what I'd consider a qualitative measuring tool. They are useful the way Phil presents them but not for comparing between different charts.

Before I continue though, I need to introduce an ahem technical term called "painterly effect" or smudging effect. This is the result to too-agressive noise reduction, making the photo look like a 1900's century Seurat aka a smudged or pointilist painted look. To "make" one of these yourself, go to photoshop, take a file, and apply filter-dust and scratches and crank it up.

One thing I'm beginning to understand from using various cameras and from learning from writings by y'all and Thom Hogan is that noise reduction and the way noise reduction is performed is very important. I'm coming from using what is considered by most to be a VERY noisy camera (Kodak) to what is considered a VERY clean camera (20D and MkII).

However, the way the two companies (Kodak and Canon) handle the noise has a lot to do with this reputation. Canon cameras do agressive noise reduction on the hardware, Kodak (and Nikon) cameras do NOT do any noise reduction on the hardware, saving it for later processing in firmware, or Nikon Capture, Photoshop, etc.

The end result is what matters though, right? With the Kodak, I could start with a noisy RAW file, + a totally unacceptable JPEG and end up with a kick butt shot that didn't look too "painterly" Here's an example of a shot that people just would NOT expect out of the Kodak:

This shot was taken handheld at f3.3@ 1/20th ISO160.

Here's a crop from the lower left corner:

The JPEG was totally unacceptable, but with some good noise reduction applied, it comes out well. I agree it's not perfect, it's a little blurry, but I would have to say (seeing the FF Canon lens thread) that's just about as good as it's going to get folks. That's just my opinion of course.

With a Canon camera, I could have taken the same shot, and the noise reduction would have been applied before the RAW file was written to the card. The shot would look great. But I see this as a good and bad thing. It means the camera has good built in noise reduction, but it also means that the control of noise has been taken away from the user. With most people and in most cases, this is fine and even desirable, but not all the time. In talking with 1DmkII users lately, some report that their images are just too smudged or too smoothe looking, making them appear fake. I don't know if this is a legacy from the film shooting days, but I've heard it from more than one person. I tend to agree that sometimes, some noise looks good, but it depends a lot on the characteristics of the noise. I haven't had a chance to use my Canon cameras enough to say " I love the Canon noise reduction and feel like I don't need to control noise myself when using this camera." I'm hoping I will be able to say that though, so like with so many other things (metering, flash, etc) I can just "trust" the camera and electronics.

I know we haven't discussed it here before but I'm sure a few have thought about it but noise looks different from different sensors. Some is easier to remove than others and some looks better than others. End result noise may not vary a lot for the same sensor, but REALLY varies by camera brand depending on firmware - this has been proven by the 8 megapixel consumer cameras which all use the same chip.

So what kind of noise control and noise do you like? This will be especially important for folks looking to buy the D2x. Nikon has already said that they will not be using agressive on-chip noise reduction.

Cheers

James

Thanks James,

As you are someone who has owned and shot DSLRs from Canon, Nikon, Fuji and Kodak - your opinions are very valuable on discussions like these. I find that I tend to dislike most camera kit I try as opposed to own because you never have it long enough to get as used to it as your normal camera - and therefore tend to dislike the things it does differently.

Alex

Here's a quote from Thom Hogan that is very relevant to this discussion:

Noise is pretty much a subjective issue, not objective. Yes, you can "measure" Std Dev values and get a vague sense of the noise production, but it says nothing about the look. Now, some folk don't like the D2h's grain-like and visible noise, some people don't like the Canon's processed noise reduction, which cleans up tonal ramps but tends to put a falseness on edge detail. But I've yet to see a camera that doesn't have significant visual impact when used at high ISO values. Choose your poisen, but make sure you know what you're choosing and why.

Cheers

James

I think you said it right, Herb.  There's a reason why selectable ISO is offered (I said programmable before) and if it's not to minimize the impact of noise I can't imagine what it would be.  We should see some straight lines on the ISO noise graphs if there wasn't some significant noise reduction at higher ISO's going on.  I just think your conclusions backwards from that.  How can a result from a high ISO measurement tell you anything about base ISO performance when the signal processing is not the same?

The reason you don't see straight lines is that the relation between exposure and noise is not linear, and x-axis is a log scale. I've plug in some numbers and alot of them actuall fit the model pretty well. That is noise with a fixed part and part that depends on the square root of the exposure. Much of noise reduction that Thom keep talking about by Canon is just circuit techniques that calibrates out the pixel mismatchs (the fixed pattern noise) in CMOS.

The high ISO performance tells you how good the low ISO performance can be at lower exposures

I'm still grappling pretty unsucessfully with this one.

While I can see that the high ISO performance CAN tell you about the intrinsic signal to noise characteristics of the sensor. The high ISO performance is also affected by noise reduction processing by the camera at those higher ISOs. When such processing is applied the high ISO performance cannot tell you everything about the low ISO ability to record shadow detail (where such NR is not applied to the same degree).

Looking at the graph again. The Nikon D2H and Canon 1D Mk2 have very similar noise performance at ISO 3200. So you might expect them to have similarly low noise at lower ISO. Yet the Canon is much better.

Alex

The reason you don't see straight lines is that the relation between exposure and noise is not linear, and x-axis is a log scale. I've plug in some numbers and alot of them actuall fit the model pretty well. That is noise with a fixed part and part that depends on the square root of the exposure. Much of noise reduction that Thom keep talking about by Canon is just circuit techniques that calibrates out the pixel mismatchs (the fixed pattern noise) in CMOS.

But since Phil doesn't say exactly what the test is, I can't assume what the Y-axis is. That was my whole point, and I've seen some pretty straight lines in other reviews, this one for example, so I'm not convinced. Clearly, noise reduction has a profound impact on the shape of these graphs. Otherwise, they'd all have the same curve whether it's a straight line or not.

What's the model you are using then and how can you be sure it fits with Phil's? It's impressive that you can confirm Phil's results when he doesn't describe how be achieves them.

The high ISO performance tells you how good the low ISO performance can be at lower exposures

Well, you can say that Herb but that doesn't make it so. Why is it that the Canon 1D* series cameras have worse ISO 50 performance than ISO 100? Everyone but Phil claims that's the case.

As I said before, altering ISO changes the hardware noise reduction techniques as is proven in the charts. As a result, any measurement at high ISO has no bearing on the performance of the camera at low ISO. You need to measure that using low exposures at the low ISO directly. Garbage in, garbage out.