Product Watch: EOS-1D Mark III dSLR

[Riley]

Advertisement (gone after registration)

One stop? Did you look at the link I posted? ISO 6400 looks like M8 640!

 

Canon cmos's advantage has very little do to with fill-factors and much to do with the noise reduction circuity done on a per-pixel basis. In fact, there are three or four transistors on each pixel that subtract out noise. No mush there. Extremely high quality detailed high ISO. Ironically, cameras that use a CCD usually are the ones using mush generating noise reduction post ADC. Like the M8.

 

it might look like M8 at 640 in your house, other chips do better

on that basis the last time I looked iso 6400 - iso 3200 was 1 stop

 

I looked at the frames presented, they are far far from sharp, so where did the sharpness go? If your read Canon's patent literature, you find that the main purpose of the DIGIC chip is to improve the native defects of the CMOS sensor, ie trade noise for mush

[deltoid1]

it might look like M8 at 640 in your house, other chips do better

on that basis the last time I looked iso 6400 - iso 3200 was 1 stop

 

I looked at the frames presented, they are far far from sharp, so where did the sharpness go? If your read Canon's patent literature, you find that the main purpose of the DIGIC chip is to improve the native defects of the CMOS sensor, ie trade noise for mush

 

You seem to be the only person so far that thinks those samples are soft. On dpreview, ALL of the forums are marveling at those samples.

 

You are confused as to how the Canon cmos chips lower noise. A link for you perusal:

 

CANON’S FULL-FRAME CMOS SENSORS: THE FINEST TOOLS FOR DIGITAL PHOTOGRAPHY

[Riley]

im not much interested what goes on over there, i doubt anyone else is either

what i do notice about the white paper you present, it is titled

 

CANON’S FULL-FRAME

CMOS SENSORS:

THE FINEST TOOLS

FOR DIGITAL

PHOTOGRAPHY

 

so now its FF as well is it ?

the relevance is then what ?

the source is what ?

 

i would just say again

looking at the iso 3200 file taken with flash, and failing light outside

i do not find it an impressive leap in iso performance

look at the window at left where the flash didnt reach

that isnt clean at all is it ?

look at the lady at right, her hair has little detail, wonder where that went

 

to me it simply equates to adding another stop to iso re 6400

no matter the consequences

 

cheers sport

[macusque]

You seem to be the only person so far that thinks those samples are soft. On dpreview, ALL of the forums are marveling at those samples.

 

Make it two.

Those samples ARE soft, period.

I don't care what dpreview Canon users rave about, just look with your eyes. Even after good USM they lack fine details.

Maybe they are misfocussed, maybe it was a bad lens, but those samples can't be seen as an example of sharp output. I'm waiting to see more.

 

Anyway Carmen, I totally agree on the 6400 ISO noise, it's much more than 1 stop better, even compared with the excellent 5D I'd say it's two stops better, which is a very impressive improvement.

[carstenw]

Carmen, I also think those samples are soft. Typically Canon, and these samples look even worse than my 5D did. The high ISO images look very clean, but until the rest of the image quality is in place, it is pretty academic, isn't it. You can get very clean high ISO pictures with an M8 and Noise Ninja, but you will lose sharpness, just like this. Let's wait for some better samples.

[jaapv]

You seem to be the only person so far that thinks those samples are soft. On dpreview, ALL of the forums are marveling at those samples.

 

You are confused as to how the Canon cmos chips lower noise. A link for you perusal:

 

CANON’S FULL-FRAME CMOS SENSORS: THE FINEST TOOLS FOR DIGITAL PHOTOGRAPHY

 

 

I must agree with Riley. I'm sure this camera can do better. And I would not, personally, hold out the majority of posters on DPReview as the pinnacle of photographic wisdom . Having owned a number of Canon DSLR's myself, I can only say that image quality is very high,high ISO is smooooooth, but that the images to get boring over time, I call it "ironed flat" some other poster called it "plastic fantastic". It is all a matter of taste nowadays. I prefer the look of CCD sensors over that of Cmos sensors, but I can imagine this being the other way around with others. And there is always Neat Image....

[Riley]

Advertisement (gone after registration)

Carmen, I also think those samples are soft. Typically Canon, and these samples look even worse than my 5D did. The high ISO images look very clean, but until the rest of the image quality is in place, it is pretty academic, isn't it. You can get very clean high ISO pictures with an M8 and Noise Ninja, but you will lose sharpness, just like this. Let's wait for some better samples.

 

Im pretty happy with that, we sure havnt seen enough, but the point I am wishing to convey is this. IF you can establish an image that looks like this with PP from any other camera.

What then is the difference ?

 

thats what I ask, what is the difference

[gravastar]

I suggest you form an opinion by looking at these samples.

 

Canon EOS-1D Mark III Digital Camera - Hands-On Preview

 

and the review:

 

Canon EOS-1D Mark III Digital Camera - Hands-On Preview - The Imaging Resource!

 

ETA: As far as I can tell the +EV corrections listed below the images are because the scene being photographed doesn't average to an 18% (or is it 12.5% ?) gray. For example the test chart which is predominantly white has a +1.3EV correction.

 

 

Bob.

[Riley]

Ive already seen them Robert

they are sourced from here

 

Canon EOS-1D Mark III Digital Camera - Hands-On Preview

[johnll]

One reason we are talking about this camera on the Leica forum is it validates Leicas choice of the 1.3X sensor.

... except that many of the folks buying M8s are coming from film cameras, so their existing lenses don't produce the same FOVs any more. Most of them will need to get or or two wider-angle lenses than they had before.

Most people were writing of the "H" sensor as dead as a dodo because it was supposedly a 5th wheel between the full frame and the 1.5X sensor. Actually I think the 1.3X sensor hits the "sweet spot" for the sports/wildlife crowd and also for the "street" photoghrapher AKA Leica user.

That's true for sports. For wildlife, especially birds, the 1.5/1.6x sensor is probably first choice.

The full frames advantage is really only realized for extreme blowups or with tilt/shift and other more studio requirements.

As mentioned above, FF allows you to use all your old lenses the way they were meant to be used, and provides the biggest pixels for any given pixel count, and therefore also the best DR and high ISO performance at any given level of technology.

The other thing that attracts the interest of the M8 user is the high ISO achievement. I would loooove to see a varient of the Canon MkIII sensor in the M9 !! ...

What I'd really like to see in the next M would be an FF sensor (with really good DR and high ISO perf) and decent internal IR filtration. We can handle any residual vignetting (left over after the special offset microlenses have done their work) in PP.

[mjh]

What I'd really like to see in the next M would be an FF sensor (with really good DR and high ISO perf) and decent internal IR filtration. We can handle any residual vignetting (left over after the special offset microlenses have done their work) in PP.

Unfortunately we cannot. There is only so much variation in the incident angle that shifted microlens designs can handle, and Leica and Kodak have already been pushing the limits with the M8. I guess you could optimize the mircolens array of a 36 x 24 mm sensor for a single lens, but then it would perform much worse with other lenses. While PP can deal with a limited amount of vignetting, it is equivalent to increasing the ISO sensitivity and will thus lead to increasing noise levels in the corners: -1 EV can be handled without difficulty and -2 EV will still be marginally acceptable, but -3 EV? I don’t think so. A dichroitic IR cut filter in front of the sensor is out of the question; handling the inevitable colour shifts (varying with the lens, the aperture, and probably the distance) is quite impossible, partly because the aperture isn’t known and partly because of the intolerable increase of noise in the red channel. If you think that the cyan corners you get with filters in front of the lens are bad, the colour shifts caused by a filter in front of the sensor are much much worse.

[Riley]

thank goodness you are around Michael, people go blue in the face trying to fend off this continual FF obsession here.

[deltoid1]

Im pretty happy with that, we sure havnt seen enough, but the point I am wishing to convey is this. IF you can establish an image that looks like this with PP from any other camera.

What then is the difference ?

 

thats what I ask, what is the difference

 

I've been shooting with a 5D for over a year. There is NO WAY that M8 images can be processed into what I get with my 5D at ISO 1250, let alone 2500. Not a chance. The 5D preserves SO much detail.

 

I posted that whitepaper Riley so that you could learn abit about Canon cmos noise reduction being done on a per-pixel level, If you'd like to know more, scroll down to the "noise reduction" section. I don't think you understand that when you clean a pixel the Canon is doing it, ONE pixel at a time, there is no smearing of detail. This is completely the opposite of the incamera detail smearing being done by the M8 jpeg engine and all other CCD incamera jpeg engines. 5D raw images have WAY more

detail than any CCD raw image that I have seen. And I think I have seen them all.

 

Cheers sport.

[mjh]

I don't think you understand that when you clean a pixel the Canon is doing it, ONE pixel at a time, there is no smearing of detail. This is completely the opposite of the incamera detail smearing being done by the M8 jpeg engine and all other CCD incamera jpeg engines.

Well, CMOS sensors are more noisy than CCDs to begin with, so Canon had to do something about this to make CMOS technology a viable alternative to CCDs. Prior to the Canon EOS D30, CMOS imagers were confined to dirt-cheap digicams or webcams – they were cheap to produce, but notorious for delivering poor image quality.

 

Indeed, Canon has been quite successful in reducing CMOS noise. They are using the on-chip circuitry you mentioned, although that isn’t the whole story. For example, when the device-dependent RGB values of the sensor are converted to RGB values within a standard colour space, there is a trade-off between colour rendition and noise – optimizing colour tends to increase noise, and you can get away with less noise by sacrificing a bit of the colour fidelity that was possible otherwise. And then there is the usual in-camera noise suppression that all cameras apply, including the Canon EOS models. It has to be acknowledged, however, that Canon is quite conservative in this respect; the High ISO noise suppression feature of the EOS-1 D Mark III is even turned off by default. Nikon’s DSLRs, for example, suppress chroma noise quite aggressively, to the extent that chroma noise levels are nearly constant from the lowest to the highest ISO levels, even when that entails a corresponding loss in detail.

[johnll]

Unfortunately we cannot. There is only so much variation in the incident angle that shifted microlens designs can handle, and Leica and Kodak have already been pushing the limits with the M8. I guess you could optimize the mircolens array of a 36 x 24 mm sensor for a single lens, but then it would perform much worse with other lenses. While PP can deal with a limited amount of vignetting, it is equivalent to increasing the ISO sensitivity and will thus lead to increasing noise levels in the corners: -1 EV can be handled without difficulty and -2 EV will still be marginally acceptable, but -3 EV? I don’t think so. A dichroitic IR cut filter in front of the sensor is out of the question; handling the inevitable colour shifts (varying with the lens, the aperture, and probably the distance) is quite impossible, partly because the aperture isn’t known and partly because of the intolerable increase of noise in the red channel. If you think that the cyan corners you get with filters in front of the lens are bad, the colour shifts caused by a filter in front of the sensor are much much worse.

You may well be right, at least for now. Hopefully, Leica and Kodak may push the limits a bit further in the future. Maybe some non-dichroic IR-cut filter will be possible someday? Maybe with a high refractive index / low dispersion glass in front of the sensor it may be possible to straighten out the angles of incidence somewhat? I don't know, but I hope that some day these things may come to pass.

[Jamie Roberts]

I've been shooting with a 5D for over a year. There is NO WAY that M8 images can be processed into what I get with my 5D at ISO 1250, let alone 2500. Not a chance. The 5D preserves SO much detail.

{snipped}.

 

Carmen--I've been shooting the 1d2, 1ds2 and 5d since they came out and believe me, the easy way to make your M8 look like the 5d at ISO 1250 is to shoot RAW at 640 and push the thing a couple of stops in post.

 

I can push the 640 M8 ISO this way in C1 "two" stops so about ISO 2500, with nice results. Of course, you have to do this manually; if you properly expose for ISO 640 you will blow out the results at 1250.

 

Perhaps someone here can explain why this happens; to me it shouldn't be possible to underexpose that much then recover in post without a lot of artifacts.

 

I know you can push the 5d / 1ds2 even higher in post, but then you *are* generating mush in the shadows. I have 10k plus shots that prove it in very dark venues Without exceptional exposure skills, using a Canon at 3200 isn't all that easy either (though it's well-nigh impossible with other digicams, I admit).

 

Anyway, if you've got an M8 (and I just don't remember if you do), try pushing those files around. I'm going to say this again: I know it sounds nuts, but it definitely works for me. In my brief testing, pushing the M8 in post at 640 gives me better / less noisy results than shooting at ISO 1250, too (though I think Leica is still tweaking this, personally).

 

FWIW, the Canon sample 1d3 shots I've seen so far on the Canon site *are* soft (and awful in other ways, but this, as others have mentioned, is evidently typical Canon and probably early shots too). If you own the 35 1.4L Canon you know what I'm talking about here.

 

YMMV

[carstenw]

There is only so much variation in the incident angle that shifted microlens designs can handle, and Leica and Kodak have already been pushing the limits with the M8. I guess you could optimize the mircolens array of a 36 x 24 mm sensor for a single lens, but then it would perform much worse with other lenses.

 

This is certainly true, but there are at least two promising directions which don't violate this premise:

 

1) Decrease the depth of the sensor wells and move the microlenses closer.

2) Reduce the size of the inter-pixel circuitry and make the sensors wider.

 

Both of these would allow to decrease the dependency of the reading on the incident angle. I presume that there is enough headroom in one or both of these that going from 1.3x to FF is actually possible, in some near future. It may take 2, 3 or even 5 years, but we will certainly get there. In my mind, the step from 'impossible' to M8 was much larger than the next step. Leica just needs to clean up the bugs in the current design, get the R line back on the road, and then focus on the next generation of M.

[Bertl]

Carmen--I've been shooting the 1d2, 1ds2 and 5d since they came out and believe me, the easy way to make your M8 look like the 5d at ISO 1250 is to shoot RAW at 640 and push the thing a couple of stops in post.

 

I can push the 640 M8 ISO this way in C1 "two" stops so about ISO 2500, with nice results. Of course, you have to do this manually; if you properly expose for ISO 640 you will blow out the results at 1250.

 

Perhaps someone here can explain why this happens; to me it shouldn't be possible to underexpose that much then recover in post without a lot of artifacts.

 

I know you can push the 5d / 1ds2 even higher in post, but then you *are* generating mush in the shadows. I have 10k plus shots that prove it in very dark venues Without exceptional exposure skills, using a Canon at 3200 isn't all that easy either (though it's well-nigh impossible with other digicams, I admit).

 

Anyway, if you've got an M8 (and I just don't remember if you do), try pushing those files around. I'm going to say this again: I know it sounds nuts, but it definitely works for me. In my brief testing, pushing the M8 in post at 640 gives me better / less noisy results than shooting at ISO 1250, too (though I think Leica is still tweaking this, personally).

 

FWIW, the Canon sample 1d3 shots I've seen so far on the Canon site *are* soft (and awful in other ways, but this, as others have mentioned, is evidently typical Canon and probably early shots too). If you own the 35 1.4L Canon you know what I'm talking about here.

 

YMMV

You are right this procedure works much better then shooting with 1250 or 2500 ISO.

Bertram

[Riley]

one other thing as regards M8

Sean was talking about pushing beyond the set iso or overexposing,

this because of the accuracy/or lack of it, of the iso metering at the top of the iso range

and the coincident property that tending to overexpose reduces apparent noise

[deltoid1]

This is certainly true, but there are at least two promising directions which don't violate this premise:

 

1) Decrease the depth of the sensor wells and move the microlenses closer.

2) Reduce the size of the inter-pixel circuitry and make the sensors wider.

 

Both of these would allow to decrease the dependency of the reading on the incident angle. I presume that there is enough headroom in one or both of these that going from 1.3x to FF is actually possible, in some near future. It may take 2, 3 or even 5 years, but we will certainly get there. In my mind, the step from 'impossible' to M8 was much larger than the next step. Leica just needs to clean up the bugs in the current design, get the R line back on the road, and then focus on the next generation of M.

 

The solution. Fiber optics.

[Advertisement]

Hallo,

Look here: Interesting Leica gear