Ad
Ad
Ad
Pages: « 1 ... 9 10 [11]   Bottom of Page
Print
Author Topic: Sigma SD1 review  (Read 42821 times)
Ray
Sr. Member
****
Offline Offline

Posts: 8845


« Reply #200 on: August 02, 2011, 07:42:04 PM »
ReplyReply

I agree with Michael here. There are two broad processes; capturing the image, and processing the image.

Sometimes people find that a particular type of processing using a particular RAW converter produces a more pleasing result, then attribute such pleasing results to the characteristics of the camera.

The other issue, of course, is the necessity of comparing identical scenes using 'effectively' identical camera settings and 'effectively' equal lens quality when trying to examine subtle differences between cameras, if one wishes to be truly objective.

The above comparison images are too small to allow resolution comparisons, but it should also be borne in mind that the SD14 is only 4.6mp (spatially) whereas as the D90 is 12.3mp.
Logged
Dave Millier
Full Member
***
Offline Offline

Posts: 118


WWW
« Reply #201 on: August 03, 2011, 06:18:07 AM »
ReplyReply

Resolution will be pretty similar, that particular chip is a pretty good match for most Bayer chips of 10-12MP.

I agree that the differences here are primarily processing. But even if you equalise this, there will be Foveon fans who will claim the Sigma image still looks to have better colour, is more "natural" and "3D". I'm open minded about this, but personally I haven't seen these properties in my own Sigma images yet. 

One thing that isn't talked about that much is that Foveon chip has one unique property- it has equal resolution in all colours, while Bayer resolution varies all over the place in different parts of the subject, especially in regions of saturated colour. Mike Chaney (ddisoft) noted this some years ago. It may not happen often but with some subjects (super saturated florals for example), Bayer sensor cameras varying resolution across different areas of subject can create a kind of "out of focus in patches" effect.


I agree with Michael here. There are two broad processes; capturing the image, and processing the image.

Sometimes people find that a particular type of processing using a particular RAW converter produces a more pleasing result, then attribute such pleasing results to the characteristics of the camera.

The other issue, of course, is the necessity of comparing identical scenes using 'effectively' identical camera settings and 'effectively' equal lens quality when trying to examine subtle differences between cameras, if one wishes to be truly objective.

The above comparison images are too small to allow resolution comparisons, but it should also be borne in mind that the SD14 is only 4.6mp (spatially) whereas as the D90 is 12.3mp.
Logged

My website and photo galleries: http://www.whisperingcat.co.uk/
Ray
Sr. Member
****
Offline Offline

Posts: 8845


« Reply #202 on: August 03, 2011, 08:14:53 PM »
ReplyReply

Resolution will be pretty similar, that particular chip is a pretty good match for most Bayer chips of 10-12MP.

I agree that the differences here are primarily processing. But even if you equalise this, there will be Foveon fans who will claim the Sigma image still looks to have better colour, is more "natural" and "3D". I'm open minded about this, but personally I haven't seen these properties in my own Sigma images yet. 

One thing that isn't talked about that much is that Foveon chip has one unique property- it has equal resolution in all colours, while Bayer resolution varies all over the place in different parts of the subject, especially in regions of saturated colour. Mike Chaney (ddisoft) noted this some years ago. It may not happen often but with some subjects (super saturated florals for example), Bayer sensor cameras varying resolution across different areas of subject can create a kind of "out of focus in patches" effect.



I find it strange that there are so few detailed comparisons of such issues on the internet and so few detailed reviews in general of the Sigma Foveon cameras. I recall my impression of the dpreview comparisons of the early Foveon sensors, years ago, which indicated the Foveon sensor was approximately equivalent to a Bayer Array of double the spatial pixel count, but I'm surprised that Dpreview and Imaging Resource haven't even reviewed the SD14, whilst a newer model in the meantime is now available with over 3x the pixel count, which seems to be an unprecedented leap in pixel count for any manufacturer.

Canon progresses more regularly in small increments, from 6mp to 8mp to 10mp to 12mp to 15mp to 18mp. Sigma progresses from 4.6mp to 15mp in one leap, with a price leap to match.

Where are the comparisons with the closest size of camera format and closest effective pixel count, such as the Canon 7D, using really long telephoto lenses of comparable quality?
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7234


WWW
« Reply #203 on: August 04, 2011, 12:17:02 AM »
ReplyReply

Ray,

What DPReview tests is probably related to sales statistics. I have considerable interest in test chart images taken with the Foveon as much data can be measured in those charts. Unfortunately, the sites using good test charts normally test those cameras that sell, and those seem to be OLP-filtered Bayer designs.

Imaging Review tested the Pentax 645D, and that camera was really impressive.

It is quite obvious that Sony will come with two 24 MP APS-C cameras end of August and Nikon is rumored  to present new cameras, too. Those cameras will also feature weaker OLP-filters. So the potential advantage of Foevon will be reduced, within a few months. Canon will follow suite for sure.

Sigma makes some impressive lenses like the 500/4.5, 300-800 and 800/5.6. So a Sigma SD1 may be interesting for the long lens shooter, but the same lenses are available on Nikon and Canon, and those systems have more options. You can buy any of those lenses + a D7000 for the price of the Sigma SD1 with money to spare for the D8000 (?).

One additional point is that much of the sharpness advantage of the SD1 is coming from the lack of the OLP filter, much suggesting that the same sharpness can be achieved using better sharpening on the OLP filtered image.

Explanation of the last statement:  The Foveon design has a small resolution advantage over the Bayer pattern, perhaps 7% (?). To that comes an OLP-filter needed to avoid aliasing. The Foveon doesn't have OLP filter, which essentiall means that it is aliasing thereby probably producing fake "microcontrast".

The OLP filter does not really reduce resolution but decreases "microcontrast". We can regain much of what has been lost in the OLP filter by sharpening with high amount and low radius and/or using deconvolution methods.

Best regards
Erik




I find it strange that there are so few detailed comparisons of such issues on the internet and so few detailed reviews in general of the Sigma Foveon cameras. I recall my impression of the dpreview comparisons of the early Foveon sensors, years ago, which indicated the Foveon sensor was approximately equivalent to a Bayer Array of double the spatial pixel count, but I'm surprised that Dpreview and Imaging Resource haven't even reviewed the SD14, whilst a newer model in the meantime is now available with over 3x the pixel count, which seems to be an unprecedented leap in pixel count for any manufacturer.

Canon progresses more regularly in small increments, from 6mp to 8mp to 10mp to 12mp to 15mp to 18mp. Sigma progresses from 4.6mp to 15mp in one leap, with a price leap to match.

Where are the comparisons with the closest size of camera format and closest effective pixel count, such as the Canon 7D, using really long telephoto lenses of comparable quality?
« Last Edit: August 04, 2011, 03:17:13 AM by ErikKaffehr » Logged

Aku Ankka
Newbie
*
Offline Offline

Posts: 25


« Reply #204 on: August 04, 2011, 03:09:41 AM »
ReplyReply

I'm made available a target for download and print for the seriously interested amongst us at:
http://www.openphotographyforums.com/forums/showthread.php?t=13217

I always hesitate to divert people to other websites (it's not nice to our host), but in this case I wanted to avoid posting the whole story again (hope Michael forgives me). Discussions about those targets, and the results people get, can take place here (again sorry Michael Wink ).


Thank you Bart for this wonderful chart.
Logged
Aku Ankka
Newbie
*
Offline Offline

Posts: 25


« Reply #205 on: August 04, 2011, 03:43:39 AM »
ReplyReply

Resolution will be pretty similar, that particular chip is a pretty good match for most Bayer chips of 10-12MP.

This of course is only your subjective guesstimate.

I used BartvanderWolf's excellent test chart (link earlier in this thread) to get (very) preliminary results for my Pentax K20D. This is, as said, preliminary, as I could not use a print, so I had to use computer screen instead (from far enough away with wide enough lens). The result was about 0,406 cycles per pixel. Nyquist is of course 0.5, so if we assume that Foveon does it to Nyquist, then Foveon would do 23% better resolution than K20D (per pixel) with Bayer and anti-alias filter, thus a 4.7Mp Foveon sensor would be worth 7.1 Mp K20D sensor pixels (4.7*1.23^2).

(I must admit, that the Foveon AA-less "beyond Nyquist" false detail may also (or may not) look pretty.)

I also did a test with the color chart, and the results unsurprisingly were very similar to above, only very slightly worse. However, the computer screen isn't exactly the best choice for this test, so I'll have to redo it as well, once I get some prints (which won't be soon though, as I don't have an inkjet).

Quote
One thing that isn't talked about that much is that Foveon chip has one unique property- it has equal resolution in all colours, while Bayer resolution varies all over the place in different parts of the subject, especially in regions of saturated colour. Mike Chaney (ddisoft) noted this some years ago. It may not happen often but with some subjects (super saturated florals for example), Bayer sensor cameras varying resolution across different areas of subject can create a kind of "out of focus in patches" effect.

Unfortunately it actually does not have equal resolution for all the colors. The top layer provides the most exact positional data, while the bottom one the least. The difference grows with large apertures and near the edges of the sensor and ought to manifest itself the most under low SNR conditions.

In addition this is not overly relevant, as in order for there to be significant resolution loss for Bayer for different colored subjects, the color spectrum would have to be of very narrow band. If one simply looks at the raw data, one'll notice that this is almost never the case, but insted most of the time there is quite a bit of data on all the channels which allows for quite nice demosaicing.

Observations from some years ago should be taken with even larger grain of salt than usually, as the demosaicing algorithms improve constantly. Anyhow, I can imagine that the "oof patch" effect is instead the result of pixel peeping area with detail that is too fine to be resolved and is instead blurred by the AA-filter.
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3404


« Reply #206 on: August 04, 2011, 05:31:12 AM »
ReplyReply

Thank you Bart for this wonderful chart.

You're welcome. Glad you also like it. It's easy to use and the results are revealing. Try defocusing a tiny bit, and see what happens to resolution. That will reveal how critical accurate focus is for achieving maximum resolution, and how thin the DOF actually is.

Cheers,
Bart
Logged
Dave Millier
Full Member
***
Offline Offline

Posts: 118


WWW
« Reply #207 on: August 04, 2011, 05:42:39 AM »
ReplyReply

OK, test charts will provide the exact numbers and therefore the ultimate limits. However, it's my experience that tests often imply bigger differences than you see in practice. For example in DPReview comparisons you often see noticeable differences between cameras in reviews that in practice would not be discernible in prints less than 5 miles wide. Review sites needs to keep providing the impression that camera X is a lot better than camera Y I guess, whilst the reality is the improvement is merely a tiny incremental change.

My practical experience of using the DP1 and SD14 alongside my 450D and 5D is that if there is any practical resolution difference, it doesn't show up in prints made at the maximum size of my R2400 printer. Similarly, the results of the comparison between the SD14 and the Kodak 14nx I published on my website (http://www.whisperingcat.co.uk/scans/sd14vs14nx.htm) showed no discernible difference in 24 x 16 inch prints when eyeballing the prints. Now I admit these field tests are not the same thing as a lab test, but I believe they are a reasonable reflection of what you get in real world photography - which surely is what matters, no?

For what it's worth, DPReview seem to agree with me as they found no extra detail from 12MP m4/3 cameras compared to the DP series Sigmas ( http://www.dpreview.com/reviews/sigmadp2/page17.asp)

ps

Mike Chaney made use of colour test charts in his SD14 vs 5D comparison and demonstrated quite clearly that Foveon resolution is constant for all colours (scoring a constant 1700), whilst the 5D was all over the place (varing between 1630 and 2000) - essentially a variable resolution camera by colour (http://www.ddisoftware.com/sd14-5d/ ). DPreview have some colour chart tests here as well http://www.dpreview.com/reviews/sigmadp1/page20.asp

Whether this translates to field conditions, I don't know.


This of course is only your subjective guesstimate.

I used BartvanderWolf's excellent test chart (link earlier in this thread) to get (very) preliminary results for my Pentax K20D. This is, as said, preliminary, as I could not use a print, so I had to use computer screen instead (from far enough away with wide enough lens). The result was about 0,406 cycles per pixel. Nyquist is of course 0.5, so if we assume that Foveon does it to Nyquist, then Foveon would do 23% better resolution than K20D (per pixel) with Bayer and anti-alias filter, thus a 4.7Mp Foveon sensor would be worth 7.1 Mp K20D sensor pixels (4.7*1.23^2).

(I must admit, that the Foveon AA-less "beyond Nyquist" false detail may also (or may not) look pretty.)

I also did a test with the color chart, and the results unsurprisingly were very similar to above, only very slightly worse. However, the computer screen isn't exactly the best choice for this test, so I'll have to redo it as well, once I get some prints (which won't be soon though, as I don't have an inkjet).

Unfortunately it actually does not have equal resolution for all the colors. The top layer provides the most exact positional data, while the bottom one the least. The difference grows with large apertures and near the edges of the sensor and ought to manifest itself the most under low SNR conditions.

In addition this is not overly relevant, as in order for there to be significant resolution loss for Bayer for different colored subjects, the color spectrum would have to be of very narrow band. If one simply looks at the raw data, one'll notice that this is almost never the case, but insted most of the time there is quite a bit of data on all the channels which allows for quite nice demosaicing.

Observations from some years ago should be taken with even larger grain of salt than usually, as the demosaicing algorithms improve constantly. Anyhow, I can imagine that the "oof patch" effect is instead the result of pixel peeping area with detail that is too fine to be resolved and is instead blurred by the AA-filter.

Logged

My website and photo galleries: http://www.whisperingcat.co.uk/
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3404


« Reply #208 on: August 04, 2011, 05:48:31 AM »
ReplyReply

I also did a test with the color chart, and the results unsurprisingly were very similar to above, only very slightly worse. However, the computer screen isn't exactly the best choice for this test, so I'll have to redo it as well, once I get some prints (which won't be soon though, as I don't have an inkjet).

It is possible to print it on different printers as well, as long as the PPI is matched to the specific output device. On a Fuji Frontier (300 PPI version) it will just produce a larger target (e.g. the Canon version of the target is 600 PPI, so a print at 300 PPI would result in a 260mm target, and would need to be shot from twice the recommended distance to compensate for the lower resolution).

It's harder than people think to make a Bayer CFA demosaicing fail. It requires a matched Red/Blue color with virtually no difference in luminosity to make it look its worst. Most demosaicing algorithms favor luminance resolution (like human vision), so with even modest response there, it can still do a pretty amazing job.

Cheers,
Bart
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


« Reply #209 on: August 19, 2011, 12:42:33 AM »
ReplyReply

Observations from some years ago should be taken with even larger grain of salt than usually, as the demosaicing algorithms improve constantly. Anyhow, I can imagine that the "oof patch" effect is instead the result of pixel peeping area with detail that is too fine to be resolved and is instead blurred by the AA-filter.
Sorry to bump an old thread. Bayer is indeed well suited for natural image aquisition, and one has to use quite specialized test-patterns to find scenes where full spatial color sampling (e.g. Foveon) really makes a difference.

I think that the (reduced) need for AA-filtering may be the biggest advantage of the Foveon chip.

-h
Logged
peterzpicts
Newbie
*
Offline Offline

Posts: 12


« Reply #210 on: August 29, 2011, 10:56:21 PM »
ReplyReply

Foveon does not reduce the need for an AA filter, it eliminates it. The worst thing you can run up against with the Foveon is stair stepping on sharp lines but this usually gets mushed out when printing or scaling. Although in the case of the SD1 it grinds things up into such smaller pieces than early Foveon sensors so most of the time its being limited by the resolving power of lenses put in front of it.
Only measurebaiters and pixel peepers worry about such minutia, photographers just stand back and watch folks gawk at their work on the wall.
On a related note Sigma has updated their SPP5.0 software to address the color issues.  Also the SD1 has a firmware upgrade that corrects several issues with IQ and improves 800+ ISO noise issues. Perhaps Sigma will loan an updated SD1 to Mike and Co. again to judge the progress.
Pete
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


« Reply #211 on: August 30, 2011, 02:12:54 AM »
ReplyReply

Foveon does not reduce the need for an AA filter, it eliminates it.
Not according to Nyquist. According to your taste and practical limitations in the equipment that you have tested, it may have eliminated it.

-h
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #212 on: August 30, 2011, 06:16:40 AM »
ReplyReply

And not according to Foveon themselves if you read their own papers on the subject.
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
ScottKennelly
Newbie
*
Offline Offline

Posts: 1


« Reply #213 on: December 09, 2011, 07:41:30 PM »
ReplyReply

Foveon does not reduce the need for an AA filter, it eliminates it. The worst thing you can run up against with the Foveon is stair stepping on sharp lines but this usually gets mushed out when printing or scaling. Although in the case of the SD1 it grinds things up into such smaller pieces than early Foveon sensors so most of the time its being limited by the resolving power of lenses put in front of it.
Only measurebaiters and pixel peepers worry about such minutia, photographers just stand back and watch folks gawk at their work on the wall.
On a related note Sigma has updated their SPP5.0 software to address the color issues.  Also the SD1 has a firmware upgrade that corrects several issues with IQ and improves 800+ ISO noise issues. Perhaps Sigma will loan an updated SD1 to Mike and Co. again to judge the progress.
Pete

I disagree. The jaggies I see in wires and similar stuff in large prints will not just be "mushed out" in photos from the new SD1. It does not even have twice the horizontal resolution of the previous cameras! Yes, it is much higher resolution, but if you previously were not completely happy with 20x30 prints from your SD14 or SD15, you will probably not be completely happy with 40x60 prints from the SD1. Jaggies WILL be visible, and people inspecting the print will realize you either printed the image from a low resolution digital file, scanned it at low resolution, or shot it with a digital camera. "Jaggies" do not show up in scans from a 4x5 camera, if the scan is done right. Lens resolution limits do not show up in the form of "jaggies" - unlike the resolution limits of Foveon sensor cameras. I've heard that files from the Canon EOS 5 D Mk II resize better than those from Sigma cameras. I would like to see an analysis of that. I have also read about landscape photographers who use 39 megapixel (and higher resolution) medium format digital cameras, because they want the best image quality they can get for their huge prints. I frankly do not have the budget to buy one, and I want the versatility of a lens system like Nikon, Canon, Sony, and Sigma provide. I plan to get a Sigma SD1, because it is KING now. While I save up my money, Canon may announce a replacement for the 5 D Mk II. If it is 32 or 36 megapixels, it will be a very hard choice to make, because there will still be advantages to the Sigma with its APS-C size sensor. Shots of birds made at 300mm will give me more detail and maybe not have to be cropped. The weight and cost of high quality, shorter focal length lenses is an advantage (more carryable and cheaper). Of course, the Canon will have live view and video capabilities (no doubt), and that is of interest to me, so there is an advantage of the Canon, as is the expected price. Still, it is image quality that will be making me buy a new camera, not video capabilities.

As for the person who wrote, "I still don't see the reason for all the excitement about Foveon, or even why people see so much "potential" in it. What, exactly, would it do better that what we have? What *could* it do better?" - I bought a Sigma SD14 to "test" the Foveon sensor. I am convinced that it shoots better quality images than any APS-C camera of its era. Today I have a Canon T1i, and while the T1i does best the SD14, it only barely wins, and that makes me think the 14 megapixel claim is close to reality (especially when shooting red objects). BTW, there are a number of 16 megapixel APS-C sensor cameras on the market today. Compare closely, and you will see that image quality varies widely. Some are better at resolving red than others. I believe the Sigma SD1 blows them all away, and it seems to easily best every other "standard" style (non-medium format) DSLR on the market. It is my opinion, based on what I've seen in my SD14 images, that the SD1 will not be bested until we see cameras that are much higher resolution than 30 megapixels. If the Canon 5 D Mk II replacement is 36 megapixels I will likely buy that camera instead, but probably not if it is only 32 megapixels . . . but I will have to see what sort of noise it produces at ISO 100, because one of the main advantages of the Foveon sensor is its very low ISO 100 noise levels. The photos look SO milky smooth, while capturing incredible detail. I LOVE that about the Foveon sensor cameras.
Logged
peterzpicts
Newbie
*
Offline Offline

Posts: 12


« Reply #214 on: December 10, 2011, 11:59:29 AM »
ReplyReply

Honestly I would prefer to not have the AA filter ever, it destroys pixel level sharpness. As far as cleaning up pixelation for super big prints, I would rather fix that in post than try to bring back sharpness, it is easier to destroy than recreate unless you are God.
I also own a D90 and it is nice but I just can't recreate the "presence" of the 4.7MP Foveon shots. Casual observers can see it but cannot quantify why it looks better to them.
Resolution wise the SD1 is a step too far for most purposes and creates a lot of overhead with huge file sizes and exposing every flaw in the lenses put in front of it.
Personally I would like to see a APS version with 2x 1080P horizontal resolution. This would weigh in at 9.7MP.  This would open up the option of 2x2 pixel binning feature to produce some really sharp DSLR video. Plus this could allow better filling of diodes with light to improve high ISO operations.
Pete
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


« Reply #215 on: December 11, 2011, 04:47:41 AM »
ReplyReply

Honestly I would prefer to not have the AA filter ever, it destroys pixel level sharpness. As far as cleaning up pixelation for super big prints, I would rather fix that in post than try to bring back sharpness, it is easier to destroy than recreate unless you are God.
Actually, some sharpness can be brought back relatively easily if you have a noise-free image in the first place, and you know (or are able to guesstimate) the blurring function. Blurring is a global function that merely reduce the level of the desired signal.

Moire/Aliasing artifacts can be really hard or impossible to fix. For stair-stepped edges you might be able to do something manually at pixel level, but if you are snapping people in a wedding with pin-stripe suits, or fabrics for a catalog, the task quickly gets very hard and/or time-consuming. Aliasing means that frequencies are falsely recorded, and you/some algorithm would have to _guess_ if it is false or real to remove it.

As sensor density increases, per-pixel blurring will move further and further into spatial frequencies where it cannot be seen unless the image is printed really large and the viewer is standing really close, or cropped really hard. Aliasing/moire, on the other hand, can negatively affect lower frequencies no matter if the camera system is capable of 100MP or 1000MP.

-h
« Last Edit: December 11, 2011, 05:52:10 AM by hjulenissen » Logged
Pages: « 1 ... 9 10 [11]   Top of Page
Print
Jump to:  

Ad
Ad
Ad