High ISO shooting

[Panopeeper]

Compare 5DII and 1Ds3 at DxOMark. Look at the Color Sensitivity detailed results. At low ISO, the 1Ds3 is better. At higher ISO's this advantage goes away, presumably because of less noise from the 5D2.

Jeff, I have seen that, however I dismissed it not ony as irrelevant, but as unreliable as well: the difference with ISO 100 is 21.9 bits vs 21.6 bits.

According to DxO, the SNR at 18% of the 5D2 is a tiny bit better than the 1Ds3, the DR measured at SNR=1 of the 5D2 is a tiny bit lower, than that of the 1Ds3, namely  11.16 EV vs 11.25 EV.

My experience with such measurements shows, that not only the measuring accuracy is the limit. Worse, two shots made by the very same camera, within seconds, may exhibit greater differences than the posted ones.

The effect of the difference of the color filters over the sensels is the same as the effect of a filter on the lens or the effect of the change in the illumination; it has nothing to do with the amplification, which occurs at a later stage.

This issue demonstrates, that one has to be cautious when evaluating such comparisons with close results.

[Panopeeper]

Often but not always: increasing quantum efficiency with no change in dark noise levels or well capacity leaves DR the same: it just moves the minimum usable ISO speed up, with the higher base ISO having about the same DR as the lower base ISO did before.

Worse still, consider one hypothetical "high ISO tuning" strategy: keep the total cell size ("pixel pitch") the same, but reduce the well size within it and so reduce full-well capacity, and then use good micro-lenses to steer all the light into the smaller wells

I suppose there are numerous examples for such development, right?

[Jeremy Payne]

Perhaps others are taking different sorts of images

Perhaps?  Here's what ISO 12,800 can do for you ...

http://chsvimg.nikon.com/products/imaging/...mg/pic_004b.jpg

[Panopeeper]

Jeremy, this is an excellent demonstration for how good a huge, horrendeously noisy image looks sufficiently downsized (and thus ISO 12800 is really useful), but would not it be better to insert a link instead of a 6.5 MB image in-line?

[dwdallam]

Perhaps?  Here's what ISO 12,800 can do for you ...

http://chsvimg.nikon.com/products/imaging/...mg/pic_004b.jpg

Anything else need to be said? lol  Excellent example!

[ejmartin]

I did not dispute this aspect. Jeff's claim was, that the change of the filters negatively affects the low ISO capability, in other words the effect depends on the ISO. All I am saying is, that there is no such connection.

Yes there is.  Or rather, there is a negative impact, which is uniform across ISO.  By decreasing the color selectivity of the CFA, in a converted image the colors will be desaturated (easy to see by taking the extreme limit of no color selectivity in the CFA, which yields a greyscale image).  Color saturation has to be restored in the conversion to the output color space, and that saturation boost increases chroma noise.   Canon seems to have made a calculation that the gain in overall photons recorded is a worthwhile tradeoff, since their "R" pixels are practically Yellow.

I'm not sure, but I seem to recall that the same color selectivity issues plague Foveon technology -- that discriminating color by the absorption depth of the photon is not so accurate, and consequently the RAW data requires a large saturation boost which dramatically affects noise performance.  IIRC there is a section of code in dcraw's Foveon processing that does median filtering of chroma channels in order to beat down that component of noise.

[Panopeeper]

Yes there is.  Or rather, there is a negative impact, which is uniform across ISO

You found a rather circumstantial way to express your agreement with me.

[JeffKohn]

You found a rather circumstantial way to express your agreement with me.

Well then I'm not sure where your disagreement with my original statement lies. What I said was that Canon took steps to improve overall high ISO performance, and that there was an negative impact on color response at low ISO. I didn't specifically say there was no impact to the color response at high ISO, although I do think that the impact is pretty much lost in other aspects of the high ISO performance, which is why you get a net gain in image quality at high ISO.

You can parse words however you want, I'm just glad Nikon made the design decisions they did with the D3x, rather than the design decisions Canon seems to be making on their newer cameras.

[Jonathan Wienke]

Jeremy, this is an excellent demonstration for how good a huge, horrendeously noisy image looks sufficiently downsized (and thus ISO 12800 is really useful), but would not it be better to insert a link instead of a 6.5 MB image in-line?

What are you talking about? 4256x2832 pixels is the original image size. Yes, there is some noise visible, but not so much it would significantly detract from the image, especially in a print.

[Jeremy Payne]

Yes, there is some noise visible, but not so much it would significantly detract from the image, especially in a print.

I printed it at 12 x 18 on premium luster and it looked pretty darn good.

Looks to me like the new 12,800 may be as good or better than 6400 was on the D3/700.

[Panopeeper]

What are you talking about?

Originally, Jeremy inserted the image into the message instead of a link to it. When you looked at the message the first time, it downloaded that huge image and displayed it in the full size. After the downloading finished, the image got reduced to a size fitting into the frame, and the reduced one looked excellent.

In the meantime Jeremy changed the reference in a link, so you don't see it that way.

[Jeremy Payne]

Originally, Jeremy inserted the image into the message instead of a link to it. When you looked at the message the first time, it downloaded that huge image and displayed it in the full size. After the downloading finished, the image got reduced to a size fitting into the frame, and the reduced one looked excellent.

In the meantime Jeremy changed the reference in a link, so you don't see it that way.

Actually, I inserted both a link and the image ... and I took out the inline image when Gabor complained it ..

I'm so used to fast internet these days I don't think twice about an image like that, but I guess some folk still have slow lines.

[dwdallam]

Actually, I inserted both a link and the image ... and I took out the inline image when Gabor complained it ..

I'm so used to fast internet these days I don't think twice about an image like that, but I guess some folk still have slow lines.

The point for me, which I overlooked because we were stuck on ISO 1600, is that you can't even shoot at ISO 12800 with the DS3 because it only goes to 3200. So, the point again is that it would be REALLY nice to have the option of 6400 and 12800. I've also printed noisy screen images at 12x18 and when printed, if again properly exposed, the noise disappears. On the other hand, shadow noise is more of a problem, but there are ways to get rid of that too.

I too like pretty much everything Nikon has done in the high ISO area. Better to have the option than not at all. That being said, I could care less if Nikon applies a more aggressive in camera noise reduction than does Canon, as long as I can get rid of it with software. Having the option is really the point, at least for me.

I would like to get back to the OPs original concern regarding high ISO. Let's say you could shoot at ISO 102,000 with NO visible noise. Would you use it or just pack in in and go home, given that was the only way to get the shot you wanted?

[Panopeeper]

the point again is that it would be REALLY nice to have the option of 6400 and 12800

What for? You are shooting raw. You have no reason to use ineffective or even fake ISOs; they only reduce the DR.

For example the 5D2 goes up to 25600, but it is fake from 6400, and already 3200 is ineffective. In other words: you would be fooling yourself when using those ISO steps.

[dwdallam]

What for? You are shooting raw. You have no reason to use ineffective or even fake ISOs; they only reduce the DR.

For example the 5D2 goes up to 25600, but it is fake from 6400, and already 3200 is ineffective. In other words: you would be fooling yourself when using those ISO steps.

Obviously I'm talking about "effective high ISOs" -like the Nikon has at 12800.

[Ray]

The point for me, which I overlooked because we were stuck on ISO 1600, is that you can't even shoot at ISO 12800 with the DS3 because it only goes to 3200. So, the point again is that it would be REALLY nice to have the option of 6400 and 12800.

Don't be deluded by impressive numbers. There's only one real ISO and that's base ISO. Everything else is a degradation. Just because a camera doesn't specify a particular high ISO, doesn't mean you can't create that ISO with underexposure at a lower ISO. This is what MFDB shooters often do. ISO 800 is no better than ISO 100 underexposed 3 stops. With Canon & Nikon, it might be equivalent to ISO 100 underexposed 1 stop, off the top of my head. (I haven't checked DXOMark   ) .

The D3s would appear to have a high ISO performance approaching the hype associated with the original D3. That shot of the bear at ISO 12,800 looks impressive because it's not overexposed. In fact, it looks underexposed to me.

[attachment=17272:D3s_ISO_...istogram.jpg]

Even with the classic 5D, remarkably clean shots can be had at ISO 3200. I hope I haven't told this story before (because there's a tendency to repeat oneself in one's old age   ), but the last time I was in Bangkok, towards the end of my stay in Thailand, I visited a particular cabaret because they allowed photography. Many of them don't.

I got a good seat near the front row and went 'hell for leather' as fast as my 580Ex would recharge. After half an hour into the show, the manager approached me and requested I refrain from using flash because it was disturbing the performers. I'm a reasonable guy, so I obliged   . But I sensed it might be impossible to get any decent shots thereafter. Nevertheless, I tried.

Should I set the 5D to ISO 1600 or ISO 3200? I figured that every shot would be underexposed because of the relatively high shutter speed required to at least partially freeze movement, so I opted for ISO 3200 because I could at least see the shot more clearly on the LCD screen. Perhaps not the right decision because there are one or two shotst that would have benefitted from ISO 1600, at the same exposure.

The following 4 shots show the full image first, followed by a 100% crop.

[attachment=17273:1577_ful_image.jpg]  [attachment=17274:1577_100__crop.jpg]

[attachment=17275:1542_ful_image.jpg]  [attachment=17276:1542_100__crop.jpg]

A D3s in this situation would have been most welcome. However, I'm still waiting for a Nikon equivalent to the Canon 24-105 F4 IS, a tremendously useful range.

[dwdallam]

That shot of the bear at ISO 12,800 looks impressive because it's not overexposed. In fact, it looks underexposed to me.

 
That's even better. ISO 12800 underexposed and still a very usable noise related image.

Try underexposing the Canon 1DS3 at ISO 1600 and see what you get. Looks like you're shooting a pile of nats.

[Ray]

That's even better. ISO 12800 underexposed and still a very usable noise related image.

Try underexposing the Canon 1DS3 at ISO 1600 and see what you get. Looks like you're shooting a pile of nats.

My shots above are all underexposed at ISO 3200. At ISO 1600 they'd be over a stop underexposed.

[Panopeeper]

Don't be deluded by impressive numbers. There's only one real ISO and that's base ISO. Everything else is a degradation

This is BS. I suggest you to read thoroughly The Source Of Noise.

Higher ISO is less noisy than lower ISO with the same exposure; however, this advantage of increasing the ISO is getting smaller and smaller with higher ISOs, up to the point, that the higher ISO creates practically the same noise as the lower step. This is 3200 with the 5D2.

[Panopeeper]

Obviously I'm talking about "effective high ISOs" -like the Nikon has at 12800.

Which Nikon? I don't know of any such, but I don't have any raw files from the D3S.