Sigma SD1 review

[Bart_van_der_Wolf]

I thought it had been fairly well established over the years that each Foveon pixel, consisting of 3 stacked sensels, is approximately equivalent to two Bayer type sensels, or slightly less, in terms of apparent resolved detail and sharpness.

Hi Ray,

NO!!! What has been established is that for Luminance most of the difference is not due to the Bayer CFA, but to the  Optival Low Pass Filter (OLPF). It's the strength of the OLPF that determines the difference for the most part, with only some 7% attributable to the Bayer CFA. Where the Foveon shines in comparison is in Chroma resolution, which for human vision is less important (unless you shoot a lot of Red on Blue subjects with nearly identical Luminance). A test chart designed specifically for that purpose can be used to determine a worst case scenario. Comparing the MTF curves for the worst case and best case scenario can be quite informative.

Real life images will have a mix of Chroma resolutions on a Bayer CFA sensor array depending on the difference in the Luminance component of the color under investigation. How Raw converters favor Luminance demosaicking versus Chroma demosaicing will also make a difference, and then there is deconvolution sharpening ..., which works very well on OLPF images.

However, until more extensive comparisons are made with this new model, we won't know for certain. It could be the case, as the pixel count of the Bayer type sensors has increased (and will contunue to increase) and their AA filters have become increasingly weaker, the resolution advantage of that lack of an AA filter in front of the Foveon sensor is accordingly reduced slightly.

Bingo! In addition, most non-Foveon sensor designs have traditionally offered (both physically and in output pixel count) larger sensel arrays than the Foveon.

The only real benefit of a Foveon image is that it responds well to upsampling, where it may hold the apparent resolution a bit better because of the extra Chroma resolution and the lack of an OLPF. In return you get a higher probability of aliasing artifacts, and a much larger Raw file than with an OLPF filtered Bayer CFA Raw of the same output pixel size, and a problematic color accuracy and noise performance.

Cheers,
Bart

[ErikKaffehr]

Hi,

Life is not just about pixels. Also, discussing the worth of pixels is not very meaningful. With more pixels you get less interpolation artifacts, like staircase effect on edges. Larger pixels will carry more MTF than smaller pixels, as MTF is higher at lower frequencies. Graeme Netress has made some calculations on this list I think, there is an advantage to the Foveon design, but less than believed.

I got the impression that Michael and Mark (?) has found the Pentax clearly superior and the Leica best of the DSLR bunch, with Sigma being better than the Alpha 900, that being quite a treat from an APS-C camera. The authors judgement is based on large prints.

One thing to point out is that those most knowledgeable about raw-conversion are least optimistic about the Foveon. It has several weak areas. Color sensivity (bad separation of color), low ISO and presumably aliasing effects. Graeme is working with raw-conversion on the RED (high end video cameras). Bart and Bill certainly know what they write about.

Unfortunately, I have seen very few testchart shots with Foevon. Would DP-review test the SD1 we would now more about how it handles aliasing, it is very visible on their test targets, at least with the Sigma DP2.

Best regards
Erik

Michael

I can only try and interpret what you wrote here:

"If we assume (and I did going into this test) that a Foveon sensor actually has comparatively about double its linear resolution (so about 30MP for the SD1) and a Bayer equipped camera such as the 645D has about 1/3rd less effective resolution than its actual pixel count (again about 30MP for the 645D), this could actually be a competitive IQ horse race."

The Sd1 has an actual 15MP sensor, the 645 a 40MP sensor. We know that Foveon and Bayer MP are not worth the same, and to come to an agree equivalence, we can make some assumptions (as you do). We can either:

- assume the Foveon is worth more than is nominal 15MP (say 2x or 30MP)

or

- assume Bayer is worth less than its nominal pixel count (say 2/3 or 30MP)

If we do the first, we are describing 30MP equivalence vs 40MP. If we do the second, we are describing 15MP vs 30MP equivalence.

The two methods don't arrive at 30MP equivalence for both and certainly shouldn't both be employed to come to a figure that suggests that the two cameras are competitive!

What you should do is say that demosacing loses about 30% of resolution and AA filtering (for those who have it) about another 20% so a camera like the A900 is only resolving something equivalent to about a 12mp Foveon.  The 645 without an AA filter probably resolves about the same as a 30MP Foveon would. The SD1 with 15MP Foveon pixels has only half the effective pixel count of the 645. Hardly a "competitive IQ horse race"....

[Aku Ankka]

It is difficult to assess the reliability of statements made by a newbie with only 5 previous posts and with no CV (curriculum vitae) listing his qualifications or prior publications. I would suggest that Michael is eminently qualified to make a subjective review of the camera. He is a very experienced photographer and has worked with all types of cameras and has extensive contacts in the photo industry whose advice he can seek. His observations are likely valid, but his technical explanations for what he sees may or may not be accurate. I will leave that to the experts with advanced degrees and experience in digital imaging.

DXO is going to evaluate this camera some time in July and we will have some technical data with which to work and they may add their own comments. For example, their comparison of the Nikon D5000 and the Canon EOS 500D demonstrates how relatively high "color blindness" requiring large matrix coefficients can degrade image quality. Does this apply to the Sigma?

Where did you learn that SD1 is gong to get the DxOMark treatment? I couldn't find such information on their site, but didn't really dig deep If true, it sounds interesting.

Anyhow, when it comes to the "color blindness" you referrerd, the Foveon chip is in totally different world compared to any of it's competition. This is because of the very poor color separation achieved by relying on the probabilistic wavelength dependent  penetration depth of photons in silicon, instead of using for purpouse designed organic pigement or dye color filters. In addition, due to the high read noise of the 3T photodiode design (because of no correlated double sampling) and surprisingly low QE especially for the reds due to excess metal in the sensor blocking light (that's 9 transistors per photosite) cripples the colors in low-signal situations (shadows or low light in general) even more. There is an interesting paper on a new Foveon sensor design in the net describing a 4T variant which would be a large improvement (due to CDS) with 4 pixels on the top layer for each pixel on the middle and bottom, but nothing indicates that SD1 has this kind of sensor.

Wikipedia has this image http://upload.wikimedia.org/wikipedia/en/a/ab/Absorption-X3.png (the one on the right was in this article, but the one in the left is actually far more informative) - it's a bit rough, but good enough for the purpouse of realizing that color is an issue with Foveon.

As I don't have a CV as you kindly pointed out, you might want to search DPReview's forums for Joseph Wisniewski's posts on Foveon sensor - he has actually made camera(s) with Foveon sensors and is of known quality (http://www.dpreview.com/members/5851432553).

[Aku Ankka]

Well, at least we have a picture of Aku Ankka here, and he seems rather authoritative to me.

You must be a paparazzi as that's a nice shot of me (or a lookalike).

Actually, Aku and Ankka are very normal names for Finns, just like Kalle and Anka are for Swedes and Paul and Anka are for Entertainers.

[Aku Ankka]

The real aspect not really addressed in the review, and one that is, or at least should be, of great interest to photographers, is dynamic range, and that's an area where the Foveon could have issues. I'd be very keen to see how it performs there.

Michael did a great job discussing the workflow and ergonomic issues, and those are the aspects we can't measure on charts - they need a competent human.

I totally agree that the review (and reviews on this site in general) had a very good workflow-ergonomics etc. part. It's just the technology part that is lacking.

Ayhow, regarding the dynamic range, it is quite modest. The main reason for this is the high noise levels, both read noise and noise from color matrixing. Even if one were to only shoot at the base ISO, the high ISO images tend to tell an interesting story, as if they are noisy relative to the competition, there is no reason for the low ISO images to not be (relative to the competition). High ISO does not increase read noise, instead it may actually lower it compared to the low ISO if the AD converter(s) are noisy (most if not all Canons for example).

The other reason is the tendency of blowing all the channels at once which is due the sensor design. With regular sensors more highlight data can be rescued as the channels tend to blow one at a time.

[Aku Ankka]

I avoided the black hole of dynamic range, simply because without a Stauffer Wedge or similar it's almost impossible to measure, and visual judgement of this parameter is notoriously unreliable.

If you mean measuring DR in general and not just this particular camera, then you are wrong. In general dynamic range is quite easy to measure and calculate. I've done it to my camera, you can do it to yours. Emil Martinec provides moderately straightforward   instructions on his website http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html - you may also be interested in reading http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html as well.

I do not know however for a fact, how this particular camera treats it's raw files and/or if it is possible to get data out of the raw-converter without any noise reduction being applied. I do suspect however that it may be impossible to get files out without NR and this would pose a problem.

[Aku Ankka]

I thought it had been fairly well established over the years that each Foveon pixel, consisting of 3 stacked sensels, is approximately equivalent to two Bayer type sensels, or slightly less, in terms of apparent resolved detail and sharpness.

Nope. This is a myth which is advocated by Sigma marketing and Sigma fans (especially in the DPR forum). Actual measurements do not support this claim, nor theory. Bart van der Wolf has an interesting web page on demosaicing.

Also, as pointed earlier in this thread by someone, the anti-alias filter is the prime culprit for reduced resolution and especially perceived sharpness and is the reason why the images should always be treated with capture sharpening.

Below are three sample images from Pentax K20D, a camera with Bayer CFA and a anti-alias fiter - one is with no sharpening at all, the second is with capture sharpening in Lightroom, and I just threw the third in there as I toyed a bit with deconvolution as it can be used to try to recreate the information smudged by the AA-filter (though for best results a good estimate of the point spread function would be needed).

[ErikKaffehr]

Hi,

A Stouffer Wedge is just about 20 box, so it is no exquisite piece of equipment. On the other hand I'd suggest that you can let DxO do their stuff, my understanding is that they are really good at it.

Dynamic range has been somewhat infected, because it seems that different people associate the term with different interpretations. My guess is that DxO figures are absolutely reliable measuring what they do measure.

There is always an issue with measurements, they may give results that contradict other observations.

Example: I tried to match two LCD screens (an Apple iMac and an Eizo Sx). According to my eyes the iMac was much brighter but accoding to my Minolta Spot Meter F they were within 0.1 stop. Than I measured with my old, analogue, Autometer II and got results corresponding to visual impression. The Minolta Spotmeter F is intended for flash and probly measures maximum intensity. The panel probably is puls width modulated, blinks rapidly and intensity is adjusted by varying the length of the blinks. So Spotmeter F picked up peak intensity and Autometer II mean intensity. Lessor learned!

Best regards
Erik

Graeme,

I avoided the black hole of dynamic range, simply because without a Stauffer Wedge or similar it's almost impossible to measure, and visual judgement of this parameter is notoriously unreliable.

Michael

[Dave Millier]

Hi Erik

I'm not making any conclusions about the technologies here. I'm just pointing out what I interpret as a bit of muddle in the review. It looks to my like Michael has worked out the two alternate ways of identifying rough pixel count equivalence and applied them both (rather than one or the other). This implies that the SD1 resolution might be closer to the 645 than is in fact the case. Obviously, he debunks this later in the review so it doesn't matter than much but I felt it worth pointing out. I've spend much frustrating time over in DPReview engaging in such debates in the past.

By the way I'm a Sd14, SD9, DP1 owner as well a 5D, 450D user, so I've lots of practical experience with comparing relative performance of these cameras!

Hi,

Life is not just about pixels. Also, discussing the worth of pixels is not very meaningful. With more pixels you get less interpolation artifacts, like staircase effect on edges. Larger pixels will carry more MTF than smaller pixels, as MTF is higher at lower frequencies. Graeme Netress has made some calculations on this list I think, there is an advantage to the Foveon design, but less than believed.

I got the impression that Michael and Mark (?) has found the Pentax clearly superior and the Leica best of the DSLR bunch, with Sigma being better than the Alpha 900, that being quite a treat from an APS-C camera. The authors judgement is based on large prints.

One thing to point out is that those most knowledgeable about raw-conversion are least optimistic about the Foveon. It has several weak areas. Color sensivity (bad separation of color), low ISO and presumably aliasing effects. Graeme is working with raw-conversion on the RED (high end video cameras). Bart and Bill certainly know what they write about.

Unfortunately, I have seen very few testchart shots with Foevon. Would DP-review test the SD1 we would now more about how it handles aliasing, it is very visible on their test targets, at least with the Sigma DP2.

Best regards
Erik

[Dave Millier]

I accidentally shot my SD14 at ISO400 the other day in full sunlight. It was quite noisy even in bright conditions. And the noise was of a different type than you normally see. Instead of the familiar fine grained red and green speckles, there was a kind of filiamentary look at quite a large scale. I would call them blotches rather than speckles but even that isn't quite right. It look a bit like the large scale maps of the universe you see, showing the distribution of matter in the visible universe. Very uneven, with denser areas of noise then voids with no noise. Very different from Bayer noise patterns.

 

I totally agree that the review (and reviews on this site in general) had a very good workflow-ergonomics etc. part. It's just the technology part that is lacking.

Ayhow, regarding the dynamic range, it is quite modest. The main reason for this is the high noise levels, both read noise and noise from color matrixing. Even if one were to only shoot at the base ISO, the high ISO images tend to tell an interesting story, as if they are noisy relative to the competition, there is no reason for the low ISO images to not be (relative to the competition). High ISO does not increase read noise, instead it may actually lower it compared to the low ISO if the AD converter(s) are noisy (most if not all Canons for example).

The other reason is the tendency of blowing all the channels at once which is due the sensor design. With regular sensors more highlight data can be rescued as the channels tend to blow one at a time.

[ndevlin]

So this is actually why I gave up on the forum here. All the undeconvoluted technobabble one cares to read, no matter how scientifically correct, means sweet fuck-all to a photographer.  

Don't get me wrong, it's ncie to have some bona fide technical experts in the house, especially when they can offer solutions to real-world problems, which they often can, but the only thing that matters is: "what can I get out of this thing?" The camera is only worth what a reasonably skilled and experienced practitioner is able to make it do.

For the easons we both said in the article, the SD1 can produce a fine image. Not quite as good as the Leica (sorry Sean, you have to sharpen images  ), at demanding applications not all that close to the 645D, and for most everything else just about one par with the A900. 

The camera makes nice pictures. They look very sharp. The colours are seriously inaccurate. The camera is of average build and design. There is no magic.

To me, the most telling thing is how little chatter there has been about the reviews of the SD1. The mass of photographer simply don't care. That's very bad news for Sigma, and largely result of the inane pricing.

It could be the case that Sigma are not particularly worried if you don't buy the SD1 because their manufacturing is currently geared for only a small output. Maybe the recent earthquake and tsunami in Japan has affected this pricing decision.

Let's suppose, for example, that Sigma is able to produce only 500 units a month, as a result of disruptions in Japan.

Ask yourselves which is better from Sigma's point of view, to sell 500 units a month at $2,000 each, resulting in a shortage of supply, long back-orders and angry customers, or to sell 300 units a month at $7,000 each, resulting in plenty of stock in all the retail outlets and warehouses, and a gradual accummulation of unsold stock which can be later sold at a substantial discount.

Nope. The economics of sensor wafer fabrication are such that th viable price depends on a good yield, which can really on be achieved after some significant production. Moreover, who is going to tool-up on a specifialized build like this for a few hundred sensors? No one, unless they are charging a fortune for the chip. I am confident that this is not the business model.

Moreover, I had to chuckle a little. While there is, indeed, a customer for everything, I seriously doubt that at this price Sigma will sell more than 300 units worldwide EVER. Not monthly, but throughout history.  In 50 years, this will be one hot item at the Westlicht Auctions :-)

- N.

[Graeme Nattress]

"means sweet fuck-all to a photographer." - well the meaning of the words might not mean much, but the talk about the visual impact of sensor design decisions directly effects exactly how the photographs you make look.

The layered photosite design of the Foveon, where it uses silicon depth as a colour filter directly impacts the colour accuracy and noise of the image, and it's dynamic range and high ISO performance and imperviousness to chroma moire. It's lack of OLPF directly impacts the perception of fine detail, and luma aliasing performance. I'd say all those parameters mean a lot to the image, and hence to the photographer taking that image. If you don't understand the pros and cons of a camera design, how can you begin to go about getting the best out of the camera for any given situation?

It's just like Michaels discussion of the ergonomics and workflow - they directly impact how the camera is used to take photos, and how the final image is made from what the camera captured. They might not be geeky math issues, but they're technical (rather than artistic) issues that directly impact upon the image.

One of the nicest, to me, things about photography is it is a hobby that combines the technical and the artistic. By understanding the technical you can work around artistic issues, or utilize a camera in such a way as to enable creative possibilities. Similarly, a good creative decision can render certain technical aspects irrelevant to the final image.

So if the geekyness of the why are the colours "seriously inaccurate" that's fine, but the take-away point from the discussion is that it's inherent to the sensor design choices, and not something that's going to get better with a better profile. Again, it's say that's rather important to know.

So indeed, "babble" can mean little to a photograph, but that doesn't mean such a discussion can not be enlightening as to the whys and wherefores of how an image looks the way it does, and what can, and cannot be done to avoid real imaging issues.

Graeme

[Alan Goldhammer]

To me, the most telling thing is how little chatter there has been about the reviews of the SD1. The mass of photographer simply don't care. That's very bad news for Sigma, and largely result of the inane pricing.
- N.

I think all of us read the review and found it quite informative.  Some of us are scientists and are interested in the physics behind the sensor and what is required for the processing pathway in order to get good images.  I wouldn't be so harsh on us techno-geeks.  I posted early on about the shocking price for this camera, even before the review was posted.  Of course the proof is really in the images and whether they are better/worse than competitive products.  I've found the tech discussion extremely interesting as I was not familiar with the Foveon sensor before and it has sparked my interest to read a lot more about it.  Whether Sigma will continue to work on this sensor or not will largely be a marketplace response.  At this pricing point it does not look promising.

[feppe]

You must be a paparazzi as that's a nice shot of me (or a lookalike).

Actually, Aku and Ankka are very normal names for Finns, just like Kalle and Anka are for Swedes and Paul and Anka are for Entertainers.

I bet you even live in Ankkalinna (Duckburg)...

[John Camp]

So this is actually why I gave up on the forum here. All the undeconvoluted technobabble one cares to read, no matter how scientifically correct, means sweet fuck-all to a photographer.  

Don't get me wrong, it's ncie to have some bona fide technical experts in the house, especially when they can offer solutions to real-world problems, which they often can, but the only thing that matters is: "what can I get out of this thing?" The camera is only worth what a reasonably skilled and experienced practitioner is able to make it do.

For the easons we both said in the article, the SD1 can produce a fine image. Not quite as good as the Leica (sorry Sean, you have to sharpen images  ), at demanding applications not all that close to the 645D, and for most everything else just about one par with the A900. 

The camera makes nice pictures. They look very sharp. The colours are seriously inaccurate. The camera is of average build and design. There is no magic.

I think all of us read the review and found it quite informative.  Some of us are scientists and are interested in the physics behind the sensor and what is required for the processing pathway in order to get good images.  I wouldn't be so harsh on us techno-geeks.

One reason there's so much talk of tests and technology is because they're easy to talk about, if you want to take the time to learn the language and the math. Most things are defined. Unfortunately, much of the stuff that Michael talks about in his review *isn't* easy to talk about, because sometimes it seems we don't even have a proper language for it, much less any agreed-upon definitions. Again, unfortunately, that's the stuff that's most important to some of us -- much more important than most of the techno-talk, unless your major interest is in engineering as opposed to image-making (not that there'd be anything wrong with that.)

But the only thing *I'm* personally curious about with the Foveon photos is their peculiar "look." As I've mentioned a couple of times, they look like watercolors to me. Is that because of Michael's post-processing, because of the particular subject matter he's been working with lately (countryside farm shots, a traditional watercolor subject) or is it the native "look" of the sensor? Is this look a result of what ndevlin calls "seriously inaccurate" colors? Are the colors really seriously inaccurate, and how? It seems to me (looking at the photos on my monitor, which may be different than anyone else's monitor) that they are not so much inaccurate, as muted -- to use painters' language, the hue is correct but the chroma is lower than on other cameras. Are the Bayer cameras simply the Velvia of digital photography, while the SD1 is the Provia? That would be different than "inaccurate," in that it would reflect a knowing choice, rather than an error.

[deejjjaaaa]

But the only thing *I'm* personally curious about with the Foveon photos is their peculiar "look." As I've mentioned a couple of times, they look like watercolors to me. Is that because of Michael's post-processing, because of the particular subject matter he's been working with lately (countryside farm shots, a traditional watercolor subject) or is it the native "look" of the sensor?

so why don't you just look through photos of people who were using Sigma cameras professionally for years instead of MR who albeit assisted and being a photographer himself still did not use that particular equipment and related postprocessing a lot ?

start w/ such popular destination as

www.x3magazine.com/
http://vstudio-magazine.de
www.rytterfalk.com
http://www.pbase.com/sigmadslr

[Aku Ankka]

I accidentally shot my SD14 at ISO400 the other day in full sunlight. It was quite noisy even in bright conditions. And the noise was of a different type than you normally see. Instead of the familiar fine grained red and green speckles, there was a kind of filiamentary look at quite a large scale. I would call them blotches rather than speckles but even that isn't quite right. It look a bit like the large scale maps of the universe you see, showing the distribution of matter in the visible universe. Very uneven, with denser areas of noise then voids with no noise. Very different from Bayer noise patterns.

Sounds like heavy noise reduction is the culprit. I wonder if you shoot in JPG or RAW? And if in RAW, maybe you should try to do conversion with all NR off (if possible with the conversion program). A trick I sometimes use when I have to use over the top noise reduction is that after the fact I add some noise to the image on purpouse to make it feel more natural.

[deejjjaaaa]

To me, the most telling thing is how little chatter there has been about the reviews of the SD1.

why there should be any chatter about SD1 reviews and of all places here ? people who are interested know what Sigma cameras are, what to expect and they can download and try raw files... what in the world MR can tell 'em new about Sigma cameras or SPP raw converter ?

[ndevlin]

The inaccuracy of the colour I have seen is in the reds.  Beautiful colour, just not what the human eye perceives.  Not sure how relevant the cause is. The only question is can it be fixed? Profiles would do it, probably. But SPP doesn't support those yet.

As for why no LR/ACR support.....let's all think whether it's in Adobe's interest to cut-out Sigma.  They support every other camera on earth, and go out of their way to get pre-release versions to testers of new cameras, so I would bet considerable sums that they are not the source of the problem. 

Lastly, I didn't mean to say that technically-oriented discussions don't have their place. I admire people who have that level of expertise in these complex technical areas.  Rather, it's that these conversations seem inordinately dominant on the LL forums lately.

Of course, because no one is actually going to buy the camera, I guess it's not really fair to talk about real world photographic experience with it    Too bad.  This technology clearly has potential. Not in the order of magnitude its evangelists suggest, but significant nonetheless. I deserves to be more than a footnote and a Harvard Business School How-Not-To case.

- N.

[kwalsh]

At the risk of getting technical 

The only question is can it be fixed? Profiles would do it, probably.

Profiles would certainly be a step in the right direction, but there is a limit since the unusual spectral response of the SD1 color channels means pretty high metameric failure which profiles can never help.

As for why no LR/ACR support.....let's all think whether it's in Adobe's interest to cut-out Sigma.  They support every other camera on earth, and go out of their way to get pre-release versions to testers of new cameras, so I would bet considerable sums that they are not the source of the problem. 

I agree.  However, the color processing for a Foveon sensor is far more involved than any Bayer sensor.  It is very possible that it is not practical for Adobe to support this type of sensor with any color fidelity at all without drastically altering their color processing.  It likely wouldn't be worth it for Adobe to do this even if Foveon was cooperative.  That, if true, I suppose could be interpreted as being Adobe's "fault".  Wild speculation of course.  But really, the color processing for this type of sensor is incredibly different from Bayer processing and that may prove to be a significant burden on any RAW processor that considers their product's color performance to be important.

And yeah, like you said, at this price the only discussions worth having are academic ones because it's not like anyone is going to buy the camera...

Ken