low iso noise A900

[markowp7]

dear all,
i am thinking about getting an a900, at the moment i am using nikon d3, d700 and hasselblad h2 with phaseone p45 and leaf aptus 75. the main idea is to complement my gear with a lightweight hi quality hi res DSLR (which unfortunately nikon has not provided yet). and the zeiss lenses for the a900 are VERY
appealing to me...
i am not in the least interested in jpg workflow at all, so for me only the raw IQ counts.
what i have seen so far by converting the few a900 raws which are floating around is the following:
1)relatively high chroma noise in high iso images (800 iso and up). in dark parts of the image it is very apparent and estethically much less appaeling
than what nikon generates (and, apparently what the canon 5Dmark II will produce). if i am not mistaken this is also (at least in part) david's conclusion.
2)much more worrying for me is the noise in low iso images (200 iso even), again in dark parts of the image. let me just reiterate that i have only seen a few
raw images (some underexposed) and there cannot be a definite conclusion out of that.
btw, i have been using Photo Raw Processor and RAW Developer to convert the a900 Raws.
i'd like to ask david (and other members of the forum who already have an a900) what their experience with LOW iso noise is? could somebody post low iso well exposed raws?
thank you in advance, peter

[Dr. Harout]

I am still waiting news for the A900 RAW processors. Beta versions are not enough.
And welcome to the forum Peter.

[markowp7]

thanks for the welcome!
as far as raw converters go, the RAW DEVELOPER is very good.
p

[David Kilpatrick]

The low ISO noise is comparable to the A700. The image is much sharper than the Nikon D3 which is very heavily processed. In fact the noise is comparable to the A700 all the way up the scale. When comparing to the D3 or the 5D MkII, be sure to downscale the image to match their sizes - the effect is most beneficial at 12 megapixels obviously, there's little difference between 21 and 25. Iridient's RAW DEVELOPER is good but emphasises the noise (with a fine structure) more than Adobe Camera Raw or Lightroom. I would suggest you try the trial version of Adobe Lightroom 2.1 Release Candidate, which can be downloaded from http://labs.adobe.com.

The images I have seen from the Canon so far all look very heavily processed with little textural definition - nice and smooth, with no visible noise at low ISOs, but also not so different from enlarging a decent sharp 12 megapixel image because there really isn't much detail present. Canon's own landscape and portrait examples are very poor. Canon owners are convinced it's because the lenses are not up to the required quality; I do not think is true. I'm using relatively low cost, simple lenses on the A900 and getting better results than many Canon users do from the latest glass on the 1Ds MkIII. I think the camera firmware is zapping the noise - not the same way that Sony did with v1 firmware in the A700, no smearing or watercolour effects, but a subtle smoothing out of textures along with noise.

David

[markowp7]

thanks david!
just one more question: i have seen your shots with the pre-release CZ 16-35mm. and the edges of the frame were HORRIBLE. i have the nikon 12-24mm and so i am spoilt with wide angles. what is your opinion on the CZ 16-35? what do you expect from it eventually?
another comment: why do you think the nikon D3 images at low iso are heavily processed?
peter

[David Kilpatrick]

Don't judge the 16-35mm on the prototype. We were not even supposed to use it and I should delete those shots. Mladen Sever (Kiklop) is convinced that they are actually redesigning the lens to force it to match the Nikon 14-24mm in performance. The prototype certainly did not.

The Nikon raw files use a different kind of processing to Sony's approach, and the strategy is to lose low-bit values where noise is most present, in the on-chip A to D conversion. Sony does some of this in the A700 and D700 process pipeline, but it's not as marked as Nikon's method. There is some loss of highlight headroom - most visible if 12-bit raw is used, 14-bit raw is much better in the Nikon cameras than 12-bit - while Sony's 12-bit is better than Nikon 12-bit, not as good as Nikon 14-bit. As both Nikon and Sony technical staff have said repeatedly, ALL raw files are subject to processing. Part of my interview (yet to be written up) with Christian Poulsen of Hasseblad has him saying the same thing - 'there is no such thing as a pure raw file, a pure raw file would be completely useless, but our raw files are closer to being pure than anyone else' - paraphrased!

David

[markowp7]

interesting point. i have heard already that canon raw files are the 'most' processed ones....
anyway, could you post some well exposed low iso raw files? so far we (or i) have only seen bad,
underexposed, blown....examples or test charts. it is really hard to judge raw image quality so far
unless i have missed some posted raws.
btw, are you expecting more CZ lenses to come up in the near future?
thanks, peter

[markowp7]

finally i went to borrow a sony alpha 900 to do some testshoots (thanks to Marc-Antonio from the Digitalstore in Vienna). all was shot in RAW, with the zeiss 24-70mm lens. .
i used the adobe lightroom release candidate as raw converter, with noise removal and sharpening turned off.
in short my main findings are:
1) there is already some noise at base iso (160-200), in zones where grey (say, darker than neutral) prevails. a simple
noise removal application takes care of it, detail remains amazing. it recalls me much more my hasselblad H2 and
leaf aptus 75 (33 megapixel) combo than, say, my nikon d3.
2)noise becomes blotchy, with color speckles at 400-800 iso. most likely it looks still ok when downsized to, say,
12mpx, but as a 24 mpx RAW image it just looks ugly.
3) 1600 iso and above is pretty much unusable (at 24mpx), unless of course there is a LOT of ambient light (but then,
why high iso?).
i do not want to contradict previous findings, which revieled reasonable to good medium-high iso behaviour of the alpha 900,
it may very well be that the noise level in term of numbers (measured in a noise analysis programm) may be contained, but esthetically
it is extremely ugly and just cannot compete with the state-of-the-art nikons at 400 iso and above. they also show noise
(physics is unbeatable after all) but it is soooo much more pleasing to look at. in fact the sony images remind me of what the
infamous kodak 14n(x) produced.
now in hindsight there may be some unfairness in this judgement, particularly since the 20.000.- USD leaf aptus 75 (and the phase one P45)
produce horrible results at 400iso. but of course the sony is a DSLR -albeit a high res one- and comparisms with other DSLR's are tempting.
peter

[David Kilpatrick]

Have you seen LaForet's Canon 5D MkII images at 1600 - and viwed them full size, including some of the midtone and darker unfocused areas where the noise shows properly?

http://www.usa.canon.com/dlc/controller ... cleID=2128

I don't think the A900 is significantly worse on overall balance of detail plus noise (if there is any issue at all, it is that the A900 does very little NR in camera). Taking a typical shot in bad conditions, from a seminar yesterday - Clare-Louise was going over a series of model shots taken earlier, with the model looking on, and obviously the lighting is completely to pot with a data projector in the background and hardly any room light. This is a full size, processed from raw with no NR at all in ACR - no chroma NR or luminance NR (normally, I would apply 50 chroma to any ISO 1600 shot from any camera). I would admit any day that the in-camera JPEGs from the A900 are unusable unless you enable Normal or High NR, they show very strong colour noise. I also know that LaForet's Canon pix are supposed to be in-camera JPEGs which are clearly far superior to the A900 in-camera output.



http://www.pbase.com/davidkilpatrick/image/103919603 for the original size 24.5 megapixel image.

Update: I checked LaForet's images yesterday (flagged up on dPreview) and I could swear that they have all been reprocessed although the Canon site says they are unprocessed in-camera JPEGs. There were some critical comments flying around about the noise as well as the sharpness (not wonderful) 12 hours ago. They look better this morning, maybe it's just the different viewing conditions on my screen, daylight versus roomlight.

David

[markowp7]

david,
canon overkills with NR, no doubt about that. however, the alpha 900 also does a lot of on-chip NR, even
if the call it 'NR off' (when you look closely at the pic you have posted you see artifacts in the images which ONLY can come from noise
reduction, like huge dark blotches in the face, maybe made a little worse by the final jpg conversion), but most likely the real sony RAW data
are more polluted. there are a lot of speckles, small red/green dots.....very hard to eliminate effectively
with 'mild' NR without sacrificing a lot of information (although the pixel count guarantees a lot of
basic info).
it seems that nikon has done a great job with the D3/D700 as a compromise of res and high iso noise. and
i am quite convinced that what i described above is the reason why we do not yet see a 24mpx D3x or D4.
also, as an afterthought, the low iso noise of the sony is somewhat disturbing.
peter

[David Kilpatrick]

The in-camera JPEG - which I won't post - is worse, an Sony must address this because even an A100 ISO 1600 JPEG can look better for chroma noise than the A900 - there is clearly something not right with the firmware for JPEG production, unless it is absolutely necessary to use strong NR, which I have been avoiding. This shot was on Normal NR, and that could mean some raw NR processing is present. NR Off images seem to me to have sharper noise grain.

But it is also processed to leave every trace of noise visible, and uses a Medium Contrast curve without any adjustment of the black level in ACR. Take the same raw file, give it ACR NR 24 luminance, 50 chroma, and drop the black level from 5 to 2 (increases shadow noise but greatly improves midtone densities); size it down to an exact match for Nikon D3/D700 in Photoshop; do not apply any web sharpening; and you get this -

http://www.pbase.com/davidkilpatrick/image/103948712

I don't how familiar you are with Nikon D3 shots under similar conditions - I'm very familiar with them! - but it really is very similar to a D3 ISO 1600 processed to similar parameters. Quite apart from that, the A900 has some qualities which the D3 lacks. One of these is exceptionally good tolerance to a wide range of lenses. I spent 20 minute on the phone today to a freelancing amateur I know and respect, who uses several systems (Leica M8, Olympus E-3, Nikon D3 and more) who is considering selling the entire Nikon kit to go with the A900 because his main interest in landscape, and he has found the 24-70mm Nikon disappointing for ultimate fine detail on the D3. I found it very good, but then got a poor sample on the D700, so the lens is not the perfection I first thought.

This shot is taken on the 28-75mm KM, and no chromatic abb correction has been applied (f/5 aperture). There is no hint of bokeh-chromatic effects, despite the adverse light on some contrast boundaries, and overall it's a very smooth and even reception of the image from this lens. I know from daily dealings with pro Canon and Nikon users that this simply can not be expected even with an EOS 5D or a D3, let alone a 1Ds MkII or MkIII; lenses are their biggest problem. You can't imagine how happy I have been to discover that the A900 works so well with so many older Minolta/KM lenses, obviously some better than others.

Don't assassinate the A900 before it has seen six months' service in the hands of a wide range of users. I showed a set of A2 size full bleed Epson 3800 A900 prints to assembled professionals on Tuesday, and while they were baffled by why anyone would use a Sony, they looked long and hard at the results and the one thing most commented on was the price of the camera - they couldn't believe it was so low, and this was coming from Canon users fully aware of the 5D MkII. Of course the 5D MkII is a full £500 ($1000) more expensive on list price than the current best retail on the A900 here in the UK, and is not on sale yet.

The one thing which put them off most was the flash shoe. They all use Pocket Wizard and similar triggers, not sync cables any more, and expect their radio triggers to fit directly.

David

[David Kilpatrick]

The Nikon D3 is unassailable at 6400-25,600 (until we know more about the Canon alternative) but to suggest that the A900 is no better than the D5D or A700 within its ISO range is nonsense. It may be no better pixel for pixel, at 100 per cent view, but there is a stack more detail present overall.

The real question is whether anyone really needs 24.5 megapixels at high ISO settings. They are normally used for sports/theatre/concert/events, typically with long lenses where the full frame is a disadvantage; and such pictures are rarely used in a way which demands more than 12 megapixels, or indeed 6. The majority end up on the web, on TV or in newspapers where 2 megapixels would be fine most of the time. If they had issued the A900 with an ISO range of 50-400 like an 'old' (2006!) Kodak full framer, I think I would be perfectly happy with that.

Because of what you shoot, I don't think you will be queuing up for an Alpha 900. Should you ever lay hands on one, you'll probably stop being so negative. But then, you've not shot with a D3 either - so in the end you have no direct, personal experience of any of this.

D3 at ISO 400 from a press/PR shot which came in today - 100 per cent clip of normal size file:

A900 clip from my ISO 1600 D3-size reduction as last shown:

Ok, just two random samples and not a strict comparison in any way - but really, the D3 is not some entirely noise-free wonder. Like the A900, it does actually display visible noise at anything over ISO 200 - and like the A900, the more detail and texture you throw at it, the less that noise becomes apparent. That's why in order to even see this type of noise, you have to clip a bit of unfocused neutral tone.

David

[David Kilpatrick]

Don, I don't think anyone has praised the D3 more than I have. I was one of the first to use one, and one of the first to see how exceptional its results were at ISO 200 let alone high ISO. Mind you, the comparison was with the Olympus E-3 which I was using alongside it.

One of the things about the best D3 files at low ISO is the rounded, three-dimensional quality they have (it's entirely different from Canon 5D). The lenses may have a lot to do with it. I am not getting the same sort of 'luscious' tonal contours from the Alpha 900 - there is something more 'dry' about the way it images.

But - I have not used a CZ lens on it. If you think I am totally happy with the A900, think again. I actually found the extra depth of field from the A700 useful, while everyone clamours to have less depth of field. No! I really need more not less, and I like the 4/3rds and smaller formats for that reason. And I'm missing the tele reach of the 70-300mm SSM on the A700 - 450mm equivalent. Not only have I lost that, I've gained a load of vignetting and distortion which was not visible on the A700. Of course, I can crop the frame - but the A700 has finer resolution at APS-C size, and overall a better image quality with v4 (pixel for pixel).

Where the A900 scores for me is focus accuracy (so far, perfect); use of my 50mm and 100mm macro lenses at their 'normal' view angles, which with the 50mm, means my copystand can be used for some commissioned work instead of a rig up of tripod and wall etc; and the feeling in ultrawide (17mm) views, which is somehow better than using 11mm on APS-C, maybe it's just the visual effect of the large finder. But it is also proving very good at ISO 800-1600 compared to the A700, once the file is sized for a typical magazine use (say 13cm wide). And I'm very pleased with the potential to return to some types of fine art and landscape work which digital has not been convincing on so far.

Fortunately I have never used flash much, so the lack of the flash and wireless flash is no big thing. But for certain jobs I do, I will have to use the A700 or by a 58 flash. More expense, not happy.

David

[markowp7]

hi david,

The in-camera JPEG - which I won't post - is worse, an Sony must address this because even an A100 ISO 1600 JPEG can look better for chroma noise than the A900 - there is clearly something not right with the firmware for JPEG production, unless it is absolutely necessary to use strong NR, which I have been avoiding. This shot was on Normal NR, and that could mean some raw NR processing is present. NR Off images seem to me to have sharper noise grain.

ok, that explains the visible traces of NR. i did my tests with NR off and, yes, the noise had more color grain. still, the images looked processed, can't tell how, but strange. as if the noise layer was put 'on top' of the image layer.


But it is also processed to leave every trace of noise visible, and uses a Medium Contrast curve without any adjustment of the black level in ACR. Take the same raw file, give it ACR NR 24 luminance, 50 chroma, and drop the black level from 5 to 2 (increases shadow noise but greatly improves midtone densities); size it down to an exact match for Nikon D3/D700 in Photoshop; do not apply any web sharpening; and you get this -

http://www.pbase.com/davidkilpatrick/image/103948712

still blotches in her phase. these are dreaded by people who write NR software. very hard to avoid them though, canon has a lot of it, nikon much much less.
I don't how familiar you are with Nikon D3 shots under similar conditions - I'm very familiar with them! - but it really is very similar to a D3 ISO 1600 processed to similar parameters


well here we disagree somewhat, but i admit, downscaling improved the image significantly.


Quite apart from that, the A900 has some qualities which the D3 lacks. One of these is exceptionally good tolerance to a wide range of lenses. I spent 20 minute on the phone today to a freelancing amateur I know and respect, who uses several systems (Leica M8, Olympus E-3, Nikon D3 and more) who is considering selling the entire Nikon kit to go with the A900 because his main interest in landscape, and he has found the 24-70mm Nikon disappointing for ultimate fine detail on the D3. I found it very good, but then got a poor sample on the D700, so the lens is not the perfection I first thought.

well, my 24-70 nikon is quite good, the 14-24 is better though. and the 24 PC lens is fantastic. i have never used minolta lenses, just realized that the CZ 24-70 produces a lot of (removable) CA. other than that it is very good.

This shot is taken on the 28-75mm KM, and no chromatic abb correction has been applied (f/5 aperture). There is no hint of bokeh-chromatic effects, despite the adverse light on some contrast boundaries, and overall it's a very smooth and even reception of the image from this lens. I know from daily dealings with pro Canon and Nikon users that this simply can not be expected even with an EOS 5D or a D3, let alone a 1Ds MkII or MkIII; lenses are their biggest problem. You can't imagine how happy I have been to discover that the A900 works so well with so many older Minolta/KM lenses, obviously some better than others.

as far as canon goes we agree totally.

Don't assassinate the A900 before it has seen six months' service in the hands of a wide range of users. I showed a set of A2 size full bleed Epson 3800 A900 prints to assembled professionals on Tuesday, and while they were baffled by why anyone would use a Sony, they looked long and hard at the results and the one thing most commented on was the price of the camera - they couldn't believe it was so low, and this was coming from Canon users fully aware of the 5D MkII. Of course the 5D MkII is a full £500 ($1000) more expensive on list price than the current best retail on the A900 here in the UK, and is not on sale yet.

all agreed too. pixel peeping is one thing printing another. and the price point of the sony is wonderful, a real steal. still, i feel (and hope) that the sensor of the alpha 900 will not make it into a future D3x (D4).
peter

The one thing which put them off most was the flash shoe. They all use Pocket Wizard and similar triggers, not sync cables any more, and expect their radio triggers to fit directly.

David

[markowp7]

sorry, the formatting of my last post was bad, i hope you can figure it out nevertheless. peter