Go Back   Steve's Digicams Forums > Digital SLR and Interchangeable Lens Cameras > Sigma dSLR

Reply
 
Thread Tools Search this Thread
Old Oct 3, 2003, 6:55 PM   #1
Junior Member
 
Join Date: Oct 2003
Posts: 18
Default Sigma SD9 internal architecture questions??

Does anyone have more info about the overall internal architecture of the Sigma SD9?

What kinda image pipe & CPU are used? How much internal RAM buffering?

Is there actually 'image pipe' hardware coming off the sensor to do some pixel-rate DSP, or is sensor data just crunched by a CPU??

Is the Sigma Raw file format spec public? This would be nice to have running on a PDA to do JPEG/TIFF conversions. (As a matter of fact, it'd be nice to have a subsuite of the software run to some extent on higher-speed StrongARM PDAs.)

I wish digicam mfgrs would allow folks to add RAM SIMM slots a la PCs so that those who really need fast shot-to-shot times can get them if necessary.


I myself would agree w/other board posters here that the Foveon sensor is not "10MPixel" or even true 6MP. But then a 6MP Bayer array is really (approx). 4-ish MP due to Bayer-to-RGB interpolation, which essentially adds a lowpass filtering behavior.

With ongoing upgrades to firmware & postprocessing, I think the SD9 will be close to 'getting it right' for overall quality, artifact reduction, etc. But Sigma now needs to figure 'user enhancements', and perhaps formulate a new model/update.
Adding a JPEG or TIFF storage option is key. This will stop this camera from being 'out in the wilderness' w/nonstandard file formats - never know when you might have to dive into Kinko's or other photo place etc. with a CF card and get prints made.



Bill Wiese
San Jose, CA
bwiese is offline   Reply With Quote
Sponsored Links
Old Oct 25, 2003, 8:20 PM   #2
Senior Member
 
Join Date: Aug 2002
Posts: 851
Default

I had read that the reason this version of the SD-9 only supports RAW is because of the horsepower (CPU wise) required to create JPG's from the data.

For that reason, I just don't see processing RAW files on a PDA. There are, however, some pretty nifty small laptops out there. Fujitsu Lifebook P series and Sony have a new laptop about the same size.

I agree, the time is now for a new foveon camera from Sigma, but both foveon and sigma seems curiously silent on the subject. Could the SD-9 be a one hit wonder?

Declan
amazingthailand is offline   Reply With Quote
Old Feb 4, 2004, 4:27 PM   #3
Junior Member
 
Join Date: Jan 2004
Posts: 20
Default

Bwiese you wrote:

"I myself would agree w/other board posters here that the Foveon sensor is not "10MPixel" or even true 6MP.

Then I dont agree with you or the other posters as I know that the Foveon X3 is a true 10.2 megapixel resolution sensor but just happens to output that resolution data in a 3.4 megapixel "size" image.
A "true 6mp" Bayer sensor has 3 million green pixels, 1.5 million red pixels and 1.5 million blue pixels, this is just a fraction of the Foveon X3's resolution which has 3.4 million green pixels, 3.4 million red pixels and 3.4 million blue pixels.
The nearest directly comparable Bayer sensored camera is the 11mp Canon EOS1Ds which outputs about the same amount of resolution but of course not equally across all 3 colour channels like the foveon.
It has 5.5 million green pixels, 2.75 million red pixels and 2.75 million blue pixels so it actually has lower red and blue resolution than the foveon.
As it has slightly higher green resolution you would think logically that it would therefore record green coloured subjects such as foliage better than the Foveon, but in practice this has been found to be incorrect.
The SD9 and SD10 images easily exceed the sharpness of the 1Ds and in fact even get close to approaching that of medium format film camera images.

http://www.sd9.org/sd9vrsmed.htm

IMO,...The test was a little unfair as the SD9 was fitted with a zoom lens which is always a compromise and of course will produce softer images than the sharper prime lens fitted to the Mamiya.

Alf B.
Alf B. is offline   Reply With Quote
Old Feb 9, 2004, 2:55 PM   #4
Junior Member
 
Join Date: Oct 2003
Posts: 18
Default

Hi, Alf..

Well, we're kinda trying to compare apples & oranges - in two different but overlapping worlds:

The Foveon sensor in SD9 has 3.34M individual sensor sites. All the color information for a given pixel is in that sensor site (for basics, never mind postprocessing sharpening/filtering algorithms that work on small regions of adjoining pixels...)

By contrast, a Bayer format sensor has rows of RGRG..., GBGB...., sensor sites. A pixel here is not a pixel unless info from surrounding 4 (or 8 or more) sites is considered. Because some array math needs to be performed on several Bayer sites to produce an R,G, and B color trio (or Y, U, and V would work as well) for a given pixel location, there is some weighted averaging - "smearing", or low-pass filtering - of data btwn Bayer pixel sites, and additional loss if extra green contour processing is performed. Depending upon information content of image (color transitions vs. color) it takes a Bayer image sensor roughly 3X the pixel (not exactly; varies w/image content) count vs the "single pixel" implementation of Foveon.

[ However do note that a Bayer sensor run in B/W mode offers excellent resolution (that is, a 10MP Bayer sensor in B/W mode offers resolution on the order of 10MP) - though full B/W performance is slightly degraded due to color filter microlenses instead of plain B/W sensor. ]

Foveon (and by extension Sigma), in order to market the new technology, needed to 'sex up' the numbers so people could make some rough "effective pixel" comparisons. It is true that a 3.3MP Foveon plays in the 10MP Bayer league

But, really, the arguing point shouldn't be that the Foveon isn't a 10MP sensor - rather, it's that a Bayer sensor of 10MP really offers a fair, effective resolution of about 3.3MP: the comparison should be done with the lower numbers (Foveon pixel site count), not the higher numbers (Bayer photodetector site count), because a "Bayer pixel" is not really a pixel


Bill Wiese
San Jose, CA
bwiese is offline   Reply With Quote
Old Feb 9, 2004, 3:09 PM   #5
Junior Member
 
Join Date: Oct 2003
Posts: 18
Default

Hi Declan..

Quote:
Originally Posted by amazingthailand
.... read that the reason this version of the SD-9 only supports RAW is because of the horsepower (CPU wise) required to create JPG's from the data.

For that reason, I just don't see processing RAW files on a PDA. There are, however, some pretty nifty small laptops ... {snip}

Declan
The Sigma SD9 has a StrongARM CPU in it which is prob running at least at 233 MHz. With proper amount of RAM buffering of raw sensor data, proper use of a realtime operating system (RTOS), and a bit of extra buffer DRAM, multiple shots could be taken while being converted to JPEG. This is the strategy for 1-3MP cheapie Asian cameras using the SoundVISION chipset, which is basically an ARM w/sensor interface, etc. and custom software.

Also, JPEG is easily and cheaply done in hardware. Sigma could've done this.

Palmtops are now coming out with 200+Mhz ARM CPUs so horsepower is not a problem for single-pass image conversion (i.e, not editing - just converting RAW to JPG)...

Bill
bwiese is offline   Reply With Quote
Old Feb 10, 2004, 2:41 AM   #6
Junior Member
 
Join Date: Jan 2004
Posts: 20
Default

Bill you wrote:

"The Foveon sensor in SD9 has 3.34M individual sensor sites. All the color information for a given pixel is in that sensor site (for basics, never mind postprocessing sharpening/filtering algorithms that work on small regions of adjoining pixels...)"

This shows that you are mis-informed and dont fully understand the Foveon X3 technology.
At each individual sensor site there are actually 3 seperate pixels not one, the outputs of the three are not combined as a single pixel at the chip itself as you suggest.

"[ However do note that a Bayer sensor run in B/W mode offers excellent resolution (that is, a 10MP Bayer sensor in B/W mode offers resolution on the order of 10MP]"

You may be correct but by saying this you seem to be admitting that a Bayer sensor run in colour does not offer "excellent resolution", but ONLY does in B&W.
Using your calculations and reasoning then surely you must see that the Foveon X3 offers "excellent resolution" in both B&W and colour, that is, it outputs 10.2 megapixels of resolution data in both B&W and colour.

So the end result is that a colour or B&W print from a 6mp Bayer DSLR will exhibit much lower resolution than the Foveon X3 DSLR, in fact around 4 megapixels less and so its images cannot be printed as large without losing quality.


Alf B.
Alf B. is offline   Reply With Quote
 
Reply


Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



All times are GMT -5. The time now is 10:10 PM.