- Well, colour me pink » »
- « « Waterproof
About those Foveon files
Filed in Sigma - June 25, 2007I’ve just about got bored with the pathetic software that Sigma supplies to convert the SD14’s raw files. To get any colour accuracy at all you have to keep exporting a file to Photoshop to have a look at it there, and then return to Sigma’s SPP and randomly make another colour change before exporting again to see if it did what you wanted. You can’t even take a white balance off one file and manually type in the Kelvin etc figures on other files as everything is expressed in terms relative to the open file. You don’t get to see the absolute information at any point.
It won’t be long before somebody releases an SD14 support update in a colour managed converter, but in the meantime, although I can’t say anything meaningful about colour accuracy, I can comment on an number of other aspects of the Foveon files.
To start with, these are small files. Regardless of what Sigma thinks, this is not a camera for large prints. People who test such things suggest that it comes in at about a 9 to 10 megapixel equivalent if print size is your major concern. It isn’t effort free either, as to just get to the native print size of a 10 mp camera you have to do some resizing. People do make big prints from these files, but any discussion about this falls down on a lack a common ground about what is acceptable print quality (minimum viewing distances and all). What I’m saying is, if print size is one of your determining factors in choosing a camera, then don’t be fooled by Sigma’s 14 megapixel claim.
No, the point of the Foveon chip is colour, and colour is what you get. I may not yet be able to colour match precise shades and light temperatures, but there is so much colour information and it is so robust that I feel confident in saying that this is the strength of the camera, as advertised.
As an aside, the Foveon flower pics that I’ve taken really stress out other parts of the imaging system. You really have to be on top of the colour space and printer gamut issue to get the most out of the information that you have collected. A regular foveonite might want to follow the Joseph Holmes path and experiment with lots of colour spaces.
Two things have struck me in particular. The first is the existence of subtle distinctions within highly saturated colours. So the fine distinctions on that super bright pink flower with the even more super bright pink line on it are recorded with clarity. The second was that, at the low ISO values that I was using, the out of focus areas are pure colour. This shows up more clearly on a print than a web version. I’ve made prints where the out of focus background has looked like it has been poured on. There is the most smooth of colour and tonal gradation and no grit. Held at 100% on an LCD screen smooth areas are prone to showing up the pixel pattern of the screen because there is no overlay from the photo. On paper, any imperfections or pattern in the base also show through alarmingly. Use only the smoothest of papers, unless you want the paper to be a part of the result.
These observations resulted in me spending a long afternoon looking at thousands of files from as many cameras as I could muster. As it happens, I have no significant number of files from any recent Nikon or Canon, so I can avoid the usual arguments between adherents of those two faiths. All the files that I had in any quantity were from non-mainstream cameras plus a few digicams. I ended up grouping the results as follows:
1. Digicams and Fourthirds cameras – in these cameras the medium is too often a large part of the message. Either noise, or smoothing, or sometimes both, make it obvious that a smallish and noise prone digital camera is in use. This isn’t a unique characteristic. Anybody using, say, Neopan 1600 will be very aware of the medium getting mixed up with the message, but unlike several of the film aesthetics, I find the digital noise or blur aesthetic unpleasing. It hasn’t stopped me using this type of cameras, but for any given result I can think, hmm, but it would be nicer on a different chip.
2. Cameras with sensors without an anti-aliasing filter – I’ve many thousands of M8 files plus a fair assortment of files from borrowed Phase One and Leica DMR gear. The defining characteristic of all of these files is detail. Subject matter gets recorded with a precision that tempts me to say etch or engrave.
3. Foveon (SD14) – this sits by itself (I’ve never seen an SD10 file). The defining characteristic being, you guessed it, colour. Colour subtlety and colour gradation exceed what I’ve seen elsewhere (bearing in mind what I’ve already said about not being able to judge colour accuracy because of Sigma’s SPP).
Note: I’m not saying that all digicams are unusably noisy, or that fourthirds cameras cannot produce fine results. Neither am I saying that an M8 doesn’t produce good colour, or an SD14 useable detail. All these are adaptable technologies and there is a significant overlap in their abilities. But I stand by the general feel of these groups of cameras as I’ve set out. They have three different aesthetics. One I’ve grown to dislike, one I like a lot, and one I can see has a lot of potential. And that’s a lot of choice before even getting to the mainstream stuff.
I understand that the technology in the SD14 isn’t quite identical to that which is going to be in the DP1, and the chip in the DP1 will also have to deal with the heat load that the live view will cause, but my experience to date suggests that the DP1 really will bring a different look to shirt pocket photography.
Without the experience that you bring to the table, I can’t see anything here that could lead me to disagree with you. Before making the obvious statement that as long as the camera does what its owner wants it to, can you say what your experience of the SD14 is in more open and dynamic situations – how does colour look away from the flower scene for instance? can it handle movement? How is it with dynamic range and noise in shadows? Is the removal of AA filtering likely in cameras other than those you mention?
June 26, 2007 @ 6:58 am
John,
I’d happily use the SD14 for general purpose photography (with the right lens of course, not the monster macro that I had). I didn’t try it at ISO 800 or up, but ISO 400 is fine (the frog is at 400). For some subjects the gain in colour definition wouldn’t be worth much.
I haven’t tested the DR because of the problem with the raw converter, but it feels OK. When I have a converter that I trust I’ll go back and look at that some more. Shadow noise hasn’t been a problem, although it does get ugly if you underexpose, so presumably there is noise in the shadows if one cared to look.
I didn’t try the focus tracking, so it depends on what you mean by movement. I’ve got some insect shots (i.e. it acquires focus quickly). A limiting factor here is the use of ‘only’ five focus points.
The buffer size (small) and write time (inordinately long) are the real limitations of the camera. And, of course, it isn’t waterproof.
June 26, 2007 @ 9:15 am
John,
I forgot…AA filters. The Kodak SLRs didn’t have an AA filter and I believe that the Mamiya body has a user removable one.
With the M8 I’ve seen one case of moire in 6 months use, but then my subjects aren’t particularly prone to causing it.
I’d have thought that Canon could sell a non-AA filtered version of their landscape cameras, but the route that they seem to have taken with the 5D is to make the filter light rather than to remove it altogether. I’ve never seen a full size 5D file so I can’t comment on how effective this is, but users really like them, so possibly this is a good compromise.
June 26, 2007 @ 9:20 am
I have a 5D and can confirm it seems to use a light AA filter. I never saw any moiré until yesterday. I had a commercial portrait session with executives and I got moiré with the fabric of a suit. Maybe I should have used a softer lens, but with the flashes I was at f/8 and it is getting hard to find a lens that’s soft at f/8.
Regarding the qualities inherent to the Foveon files, I remain partly unconvinced.
I can easily understand they can offer a wider gamut, but I fail to see the point with all the other devices already struggling.
I am not so sure the wider gamut comes from the multi-colour pixels. After the interpolation is done on each colour channel of a Bayer matrix, there is no reason that technique could not give the same gamut. What the Foveon technique brings is unparalleled resolution. However, the sensor small size and relative small number of pixels cancel any real advantage of the sensor.
I used to think the Foveon approach was the future of sensors and that everyone would go that way. What seems to be happening is the continuous improvement of the flawed Bayer technique.
June 26, 2007 @ 11:38 am
Stephane,
I don’t think that the Foveon has a wider gamut (or not especially so), it is more that it has clearer colour distinctions and more detail in single colour areas.
The point about the problem with output devices is that you have to be more careful in managing the gamut compression because there is detail to preserve whereas a Bayer file will have lost the detail in-camera (in my hypothetical example which exactly matches the strong suit of the Foveon camera).
The small number of pixels only matters if you wish to end up with an output that is physically large. I’m currently working on a botantical project which is intended for web only output. The 4.x mp of the Foveon sensor would not be disadvantaged for that (one of the reasons why I tested it). The whole mp and resolution thing is always couched in ‘larger is better’ terms, but that ‘ain’t necessarily so’.
June 26, 2007 @ 11:50 am
Colin,
Indeed, it is bound to have more resolution per sensor square millimeter in single colour areas. However, let’s count single colour pixels in an SD14 and a 5D.
Green: 4.6M for the Sigma and 6.3M for the 5D
Red and Blue: 4.6M for the Sigma and 3.2M for the 5D.
What’s the dominant colour in botanical? I’d expect green :-)
Moreover, the larger sensor size generates less enlargement of the image the lens projects.
I’d bet you should get better resolution with a 5D in all departments. Of course, for web output none of this matters. Even on common print sizes it would not matter all that much :-)
Nevertheless, I think the strength of the Bayer arrangement lies in its fundamental flaw. It is at first look ridiculous to have monochrome pixels in a colourful world, but those monochrome pixels are much easier to engineer and Bayer sensors have kept ahead of Foveon.
Now, if either Sony or Canon put their large resources in developing the Foveon concept, of course we would all benefit. It might just happen in a few years when shrinking pixels in a Bayer matrix won’t bring anything more.
June 26, 2007 @ 1:57 pm
Stephane,
I’m not defending Sigma here, I’m exploring.
let’s count single colour pixels in an SD14 and a 5D.
Green: 4.6M for the Sigma and 6.3M for the 5D
Red and Blue: 4.6M for the Sigma and 3.2M for the 5D.
What’s the dominant colour in botanical? I’d expect green :-)
Well, that’s sort of the point of doing a test isn’t it?
The real weakness of a Bayer sensor is in something like a field poppy. Greens are easy, especially because many are in fact nearly yellow which means you get quite a lot of other pixels firing too.
The straightfoward comparison of 4.6mp to 3.2mp for the red channel assumes that you want to end up with a print/output based on a 12mp total (of a theoretically perfect red subject). If what you want is something much smaller, then better to use all 4.6 than to take 3.2, interpolate up to 12 and then interpolate back down again to (say) 5.
From the available technologies, the best camera for the job depends so heavily on what the job is.
In my mind, if I want the resolution to create a large output size, the answer is Phase One (or Leaf etc), even with the Bayer limitation. But such devices bring other disadvantages.
June 26, 2007 @ 4:10 pm
Sorry if I sounded vindictive, it was not my intention.
I just fail to see the practical advantages, but all this is largely irrelevant in my opinion. I don’t think your pictures would suffer the slightest bit if you used any other DSLR.
June 26, 2007 @ 10:01 pm