SIGMA SD14 Coming Soon...

[Jacob Mason]

When Photokina starts, there will be some interesting cameras on display, some a surprise, others not so much.
You can file the following [Sigma SD14] as one of the more pleasant surprises.
DPReview Link
Sigma SD14 teaser page
For those of you who are not familiar with Sigma's unique approach to DSLR sensor technology, see: Foveon X3 Sensor technology.

[Abdulla Bastaki]

I'd have to say the old sigma cameras are PRETTTY sharp cameras.. they resolve colors and contast better than most cameras in the market.. however the images are a bit noisey on higher iso levels. the color and contrast are all thanks to the sigma's Foveon x3 sensor (it aint ure average bayer sensor).
xtra xtra read all about it...
http://www.dpreview.com/reviews/sigmasd10/

now im guessing the new sd-14 will surely kick ass... although im still a canon boy, unless this upcoming sigma camera shows less noise at iso 800/1600.

check this picture of a lion
http://www.pbase.com/ianvermeer/image/61172935

Here are some random pictures off the old SD10

http://www.pbase.com/cameras/sigma/sd10

[Wayne Morellini]

Jacob,

If anybody is going to Photokina, I would appreciate any reports on this camera and it's performance. I have been waiting for this camera for sometime.

Abdulla,

I have information on, what is likely, the senor of this camera. It has been posted on the Internet since I received it, but that was likely an accident, because it was taken down again quickly.

I can say, if it is the same senor, expect better performance. As you know, it works by different colors penetrating to different layers of the chip. But color response has overlap in real life, and the light can still interact with the wrong layer before getting to the right one (I can't remember, but I think this was the problem). But all these problems can be compensated for in design, and this is a third generation chip.


Their is another company coming out with an alternative depth related technology for the really low end, hopefully this year or so. I am still to conclude discussions, so I won't mention their name.

[Jacob Mason]

Originally Posted by Wayne Morellini

Jacob,

If anybody is going to Photokina, I would appreciate any reports on this camera and it's performance. I have been waiting for this camera for sometime.

Nobody that I know personally.
But I'll be tuned into DPreview, in fact, their coverage has already started.
With regard to the SD14, I think a lot of people are hoping for a lot of the same improvements that other photographers expect from other cameras, i.e. more res, lower noise, faster operation, lower cost.
Granted, the Sigma may not beat out the others for overall resolution, or low noise, but they'll still hang in there.
Last I heard the chip was 1.7x crop at 3x4.5MP. And the body that was being shown looked a little different than what I thought I saw in the teaser, still looked pretty straight forward.
Any other specifics you know of?

[Wayne Morellini]

Not on the camera itself, but on the possible sensor, yes.

I would like to see somebody like Panasonic, do a cheaper Lumix X3, with RAW digital HD video. But, I doubt that is going to happen, we would probably get H264 instead, as long as they can keep it above 25mb/s in h264, then we should be well. The extra bandwidth is not a problem, these are expensive cameras in the first place, an internal HDD, or multi slot storage card interface is not going to add much extra cost to those who want this feature. (please note this is a wish list for a Lumix version, it does not bear at all on the Sigma version).

There is the possibility to incorporate something lossless, based on ideas from people around here.

The problem, is that other manufacturers have Bayer, less accurate, but because it is three times smaller, it makes a lossless video data rate a lot more practical to achieve. Something like Cineform's visually lossless RAW bayer video codec, which works in between 24-32mb/s (less, for lower quality) and was developed out of our cinema camera projects here. If chip-sets like the Ambarella, h264, are truly general purpose programmable, they could do the job for $25 (but that is a big "if"). You could possibly also program Jpeg2000 in too boot. We even have people here developing lossless inter frame differential bayer codecs that would require comparatively little in an FPGA design. I think near visually lossless is achievable in 9Mb/s (Mega bits per second) around 23:1 at 24p. My own ideas extend to lossless at 9mb/s on a 4:4:4 camera using a video format, supposedly, even less accurate then Bayer (but then again that will be a technical stretch).

Will be good to see what the different manufacturers come out with.

Once again this is just wish list, I have know no details on the camera.


I am also interested in what the different manufacturers hinted at, will be doing at the show.

[Wayne Morellini]

Came across this interesting link:

http://inst.eecs.berkeley.edu/~cs39j...onX3Slides.pdf

It is a talk some years ago, that parallels modern color septation techniques in sensors to early methods in Photography, with the X3 being the natural successor to three layered film.


Thanks

Wayne.

[Peter Jefferson]

Wayne, check your email mate..

these shots are trully amazing..

[Frank Granovski]

Geez, I gave up waiting for the downloading of the teaser page.

[Wayne Morellini]

As it is slow I thought I would kick start it off again here is an interesting site that has some information in it's forums:

http://www.sigmadslr.com/
http://forums.sigmadslr.com/viewforum.php?f=16

Simple comparison between sensors:
http://forums.sigmadslr.com/viewtopic.php?t=188

Here is an interesting comparison of different lens with the same camera. The latest is that Sigma might be trying to lock people into there own lens, but the comparison makes the Canon lens look a lot better to me.

http://www.sigmadslr.com/lenses.htm
http://forums.sigmadslr.com/viewtopic.php?t=190

New interlayer pigment filtering to shape colour response:
http://forums.sigmadslr.com/viewtopic.php?t=193

Steves is also into it:
http://www.stevesforums.com/forums/v...36&forum_id=79

Here is an slightly of little test page where they seem to try to get a 8mp debayered image, reduce it to 2mpixel, then interpolate to 8mp again and compare. I don't know what they think they are proving. If they wanted to compare foveon to bayer, they would need to get a Foveon image extract a bayer pattern and then debayer and compare. The way they are doing it of course it is goign to be less quality. as they are chucking as they are averaging out pixels to 4 mpixel blocks. Bayer interpolation works by the fact that information is recorded at each pixel location, and that the human visual response overlaps big time in blue/green, and a little in red/green. So these colors reveal information about the intensity of two colours at each location, then because colours tend to remain constant while brightness changes, the colour and brightness can be more accurately determined. This is why the film camera industry felt they could trust bayer/primary colour filter enough. these are not mere guesses but calculated estimates.

http://www.sigmadslr.com/interpolation.htm

[Marvin Emms]

I can see what they are doing. Its snake oil.

They are trying to convince us that an 8Mpixel mosaic image only delivers 2Mpixel 4:4:4 quality, and that this can be interpolated back to the same quality 8Mpixel image from the 2Mpixel 4:4:4, so therefore all the demosaicing algorithm is doing is interpolating. They have also deliberatly picked an RGB mosaic camera rather than a bayer camera as they know that a bayer camera would have a much higher visible luma resolution for a given number of pixels.

This then provides the justification for taking the 4.65Mpixel 4:4:4 output from the SD14 sensor and just interpolating to 14 or 18Mpixel images.

This is just some of the worst sort of snakeoilgarbagelieslieslies.

I can see why Forveon is releasing teasers and leaking the propaganda to 3rd part sites so they retain deniability. Its because the sensor is really poor. If you take a 12Mpixel bayer image in the best possible circumstances (white and black lines) with a clever debayering alg you really can get almost 12Mpixel images out, and better than half of that average case for real images. If you take a 4.65Mpixel sensor under the best possible circumstances you can never get anything more than 4.65Mpixels worth of luma resolution. Thats why people with 4:4:4 cameras pixelshift (In exchange for lower chroma res). The idea that the real resolution is 4.65Mpixels*3 is nuts.

And they go on about 3 layer sensors being the next evolutionary step from colour film. Sure, but film has only been with us a century and is now being replaced by something better. The next evolutionary step from the oil paintings we've had for the last thousand years would be to genetically engineer a hamster and keep it in a box with an easel staring through a hole in the side. The next evolutionary step for photolithographic etching would be to do nothing to the slab, but to throw acid over the subject and scenery until they more closely resemble the block of metal. Just because two ideas share concepts does not mean they are optimal or even sane.

Feel free to gloat about the technology Forveon, but only if this is an improvement in practical results, from the talk the new sensor actually has more noise than the old one, which itself has a reputation for high noise.

Very very dissapointed by what I see.

[Wayne Morellini]

I've seen the official specs, and it is not so clear cut. If there is increased noise, it is likely in the reds, because of the infused pigment filter (which I haven't noticed in the information I have). In the old scheme the layers above the red would have absorbed not only the blue and green, but parts of the red as well, so red was lucky last. You probably could compensate for these losses by working out the amount of loss from the charges of the different layers. I imagine there might be some possibility of charge being lost to deeper layers. You then get the problem of making layers smooth and of a certain thickness, when you get down to these scales things get rough. Then there is light coming in on an angle, which will cause it to be absorbed at a different depth then light coming straight on. With this new scheme (assuming that it is true, I have official information, and I have not heard of this before) it might improve the upper layers, but make the lower layer suffer more, but still, with the other improvements it might not be much worse then previous, it could even be better, depending on the other improvements.

[Kurth Bousman]

There's been no leaked images confirmed to be coming from the sd14. So no one who knows is talking and those talking don't know. About the sd10 , I've seen reliable tests on fredmiranda that show the sd10 outperforming the canon 1ds , i.e. 11mps. So with the 400d image improvements being a real yawn , there's alot of interest in the sigma . Will it leapfrog the 10mp c&n&p&s&o offerings with their bayer sensors ? We probably won't know until sigmas' ready to show us.

[Wayne Morellini]

To get 11mp, I would only really trust RAW bayer with a very good post debayer. There is a range of quality in de-bayering. I would only trust like 100% quality jpeg if it had a very good in camera debayer.

We had a page that we used to link to in the Digital Cinema camera threads that illustrated the differences in several bayer types. I think the averaging nature of some is the biggest problem for resolutions, but in todays day an age with resolution enhancement techniques and proper debayering we can gain very good pictures.

The way to gain over bayer with this camera, is if they can implement a hardware pixel shift, where the layers are offset from each other, 4 times the resolution is not to unachievable, while nine times is possible. I have been thinking about such things with a three chip for an effective data rate of three times less than the equivalent bayer. It is tricky to achieve and retain a good pixel quality, but I am hoping for something as good as bayer.

[Marvin Emms]

Up to twice the luma resolution with pixel shift is possible provided some processing is done, but this will reduce chroma resolution.

Pixel shift is not magic, and don't fall for that whole overlapping areas argument, its just a lie. For resolution the thing that matters is the pattern formed by the center point in each pixel.

If you shift 2 layers then you end up with a resolution equivlent to an RGB mosaic camera of 3 x n pixels. Since blue contributes so little to luma resolution, this is arguably not worth doing over just moving the green.

[Wayne Morellini]

It is possible. While the luma changes much more often then the chroma. Bayer uses this to restore resolution, as the pixel value difference from its neighbours can be used to calculate brightness for the other colours, even the chroma. The trick is in intelligently debayering the image, to increase accuracy. It is helped because human vision has a huge over lap between green and red response, and a little overlap between green and blue (no overlap between red and blue, that would be why red and green are not put next to each other). Also videos 4:2:0/4:2:2 coding, but even a number in the digital film industry has gone with bayer quality. Pixel shift works in a similar way, the edge differences can be used to calculate real pixels. Intelligently, most of the time, you should be able to determine a reasonable higher resolution pixel. This doesn't mean they normally do, just that they could (if they gave you maximum quality then that reflects badly on their more expensive cameras). Like with bayer, I imagine quality varies widely between algorithms.