Reading mode: Light Dark

Sigma RAW only

  • Thread starter Thread starter Erik Magnuson
  • Start date Start date
Go to last post
If uncompressed, the calculation should use the sensor pixels not display pixels = 2304 x 1536 x 4.5 (36bits) = 15925248 bytes or 1.51875 MB

(1MB = 1024 x 1204). One more unknown factor is sometimes a camera don't use all sensor pixels.

For compressed it is harder. Faster on the fly compress like Canon is less than 2:1. According to Phil's test on Canon 1D is about 1.7:1.

Slower compression like Nikon is closer but still not 2:1. Average according to Phil is 1.85:1.
Erik Magnuson said:
Tom said:
Make that 13.0814208984375 Mb
Click to expand...
Well, they will probably implement some type of lossless
compression. However, it's not likely to give more than 2-1. So
8MB would be my guess. (A high-quality JPEG would be 1.5 - 2MB).

--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
 
Ok, going on the current camera I have in my hand, a 9 meg
RAW file is about 2.2 megs compressed in camera. We are
talking about three times that information, and at 16 bit, not
8 bit. The embedded jpg file may or may not be there in
the Sigma file, so it is hard to even guess at what the actual
file size will be. Using the G-1 as an example, at three times
the data, we are talking about 27 megs, and whatever that
compresses to. 12 bit x3 stored in 16 bit. Someone do the
math on that.
Erik Magnuson said:
Wing2 said:
If uncompressed, the calculation should use the sensor pixels not
display pixels = 2304 x 1536 x 4.5 (36bits) = 15925248 bytes or
1.51875 MB
Click to expand...
Huh? You miscounted decimals: 15_925_248 bytes = 15MB uncompressed.

--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
 
Bob Williams said:
Ok, going on the current camera I have in my hand, a 9 meg
RAW file is about 2.2 megs compressed in camera. We are
talking about three times that information, and at 16 bit, not
8 bit. The embedded jpg file may or may not be there in
the Sigma file, so it is hard to even guess at what the actual
file size will be. Using the G-1 as an example, at three times
the data, we are talking about 27 megs, and whatever that
compresses to. 12 bit x3 stored in 16 bit.
Click to expand...
Member said:
Someone do the math on that.
Click to expand...
No, because there are so many incorrect assumptions there, it's not worth it.

1. The G1 records 10 bits per pixel. So the output of the G1 sensor is 31,457,280 bits (see http://www.dpreview.com/reviews/canong1/page13.asp ) Now what we don't know is how this data is packed in the file. From the little that's been published, it's not a simple as a raster of packed 10-bit pixel values. However, just for the sake of argument, let's assume it is. That's 3840 KB bytes of input with an output size of 2792 KB (using Phil's review numbers).

2. We don't know the Foveon RAW format either. What we do "know" is that the SD9 wil output a 2268x1512 pixel image with 3 colors * 12 bits for each pixel. If we assume again it's a simple raster file of packed raw pixel values, that's 2268x1512x3x12/8 = 15,069 KB of data before compression and before the overhead of headers, thumbnail image data, etc. Again, it could be larger if they include the entire sensor output in the RAW and not just the "usable" pixels (e.g. 15,925,248 bytes)

Again, we don't know what kind of lossless compression they will use. However, we do know that RGB raster photo data rarely compresses (via zip) better than 2-1. So it's unlikely that the RAW file size will be much less than 7.5 MB. I think that's a safe lower bound. When you add back in overhead and a more realistic compression factor of 1.7 you get 8.9 MB.

In my previous post, I was giving them the benefit of the doubt and rounding down to 8MB. That's still 4x the size of a fine mode JPEG. Larger numbers for RAW file size only make the lack of the incamera JPEG more puzzling.
--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom
 
It might be helpful for some here, so I will describe this:

In my utility ABC-View Manager, you can sort your collection of pictures on a property called "Compression". It is defined as bytes per pixel. Usually, JPG pictures use somewhere from 0.1 to 0.6 bytes per pixel. Some BMP images up to 4 bytes per pixel.

Use ABC-View Manager to sort on this property, then you can directly pinpoint the worst compressed images in your collection and convert them.

Download this application (only 1395 Kb) here:
http://www.abc-view.com/abcview.html

Kind regards,
Nils Haeck
http://www.abc-view.com
 
Wrong. For a JPG file, the RGB values are first transformed into Y Cr Cb values. This is just another color space, but still has 3 values of 8 bits each! So your remark about JPG files not being 24 bit is nonsense.

Since the Y channel is most sensitive for our eyes, it is the least compressed. The Cr and Cb channels are more heavily compressed.

Perhaps you have mixed up GIF and JPG formats?

Nils
http://www.abc-view.com

By the way, I never heard of an 8 bit word. Such a thing is called a "byte".
Bob Williams said:
Wrong. A jpg captured in the camera, and available to any
program, has already been processed, and each pixel is described
by an 8 bit word for lack af an easier explaination. With a
12 bit AD, that information is condensed from all of the samples
involved, and is stored as an 8 bit word that defines the color.
There is no further interpolation required to view the file. It
is already done. All the viewer program has to do is replace the
numbers in that 8 bit word with a colored pixel. This is done
thru working color space lookup. There are not 3 eight bit words
stored for each pixel, as would be required to produce true
24 bit color.
Click to expand...
 
Erik, nice research, but it isn't accurate. The G1 does not
output 10 bits, even in RAW. It only supports 8 bit output.
When I first got it, I tried to get one of the people playing
with the Canon SDK's to provide 16 bit suport for the G1 so
that the added info could be used. They said it was impossible,
as the camera would not support the output. I just picked
three G1 RAW files at random from some holiday shots. The
three files are 2560, 2808, and 2298. My estimate may have
been a little low. 2500 might be closer to a medium size. When
opne in an editor, they are just a hair over 9 megs. This is not
going from review data, it is from months of use and thousands
of shots with the actual camera. The D30 and the G2 support
the 16 bit output option, and allow you to turn the false color
processing on or off. The G1 does not support these options.
Erik Magnuson said:
Bob Williams said:
Ok, going on the current camera I have in my hand, a 9 meg
RAW file is about 2.2 megs compressed in camera. We are
talking about three times that information, and at 16 bit, not
8 bit. The embedded jpg file may or may not be there in
the Sigma file, so it is hard to even guess at what the actual
file size will be. Using the G-1 as an example, at three times
the data, we are talking about 27 megs, and whatever that
compresses to. 12 bit x3 stored in 16 bit.
Click to expand...
Member said:
Someone do the math on that.
Click to expand...
No, because there are so many incorrect assumptions there, it's not
worth it.

1. The G1 records 10 bits per pixel. So the output of the G1 sensor
is 31,457,280 bits (see
http://www.dpreview.com/reviews/canong1/page13.asp ) Now what we
don't know is how this data is packed in the file. From the little
that's been published, it's not a simple as a raster of packed
10-bit pixel values. However, just for the sake of argument, let's
assume it is. That's 3840 KB bytes of input with an output size of
2792 KB (using Phil's review numbers).

2. We don't know the Foveon RAW format either. What we do "know" is
that the SD9 wil output a 2268x1512 pixel image with 3 colors * 12
bits for each pixel. If we assume again it's a simple raster file
of packed raw pixel values, that's 2268x1512x3x12/8 = 15,069 KB of
data before compression and before the overhead of headers,
thumbnail image data, etc. Again, it could be larger if they
include the entire sensor output in the RAW and not just the
"usable" pixels (e.g. 15,925,248 bytes)

Again, we don't know what kind of lossless compression they will
use. However, we do know that RGB raster photo data rarely
compresses (via zip) better than 2-1. So it's unlikely that the RAW
file size will be much less than 7.5 MB. I think that's a safe
lower bound. When you add back in overhead and a more realistic
compression factor of 1.7 you get 8.9 MB.

In my previous post, I was giving them the benefit of the doubt and
rounding down to 8MB. That's still 4x the size of a fine mode JPEG.
Larger numbers for RAW file size only make the lack of the incamera
JPEG more puzzling.

--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
 
Bob Williams said:
Erik, nice research, but it isn't accurate.
Click to expand...
It still does not explain the fuzzy math on why you think the RAW file size is going to be 27MB. I'm assuming they do all they can to reduce the file size (w/o losing data.) You seem to think that they will attempt to maximize it. Most people think smaller files are a good thing (again assuming no loss.)--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom
 
Maybe I don't understand, but why is not JP2 or Lurawave not considered in this case? I thought is was almost like TIFF in quality altho it can do very high compression. Can't this be put in a digicam?
Peter G said:
I am in total agreement with you on the space issue. Many people
don't even use the best JPEG mode that thier camera has in an
effort to save space.

There is a price to pay for buying first generation technology.

Peter
Wing2 said:
The RGB data out of such sensor are extremely unlikely to be linear
and 99.9% need extra processing.

At first I thought they need some kind of in camera processing or
they cannot display the image on the LCD. But then I think they
might just selected a small number of pixel to display.

No jpeg is more than big inconvenience, number of shots that can be
taken is very low. 1G microdrive only good for 60 shots? That's
quite expensive. Also expect long write time per picture. Even the
latest 24X CF can only write 3.6MB per sec. (microdrive slower?)

Even if we assume the buffer of Sigma is very big, it will take a
long time clear and cannot change CF nor shoot during such period.
Erik Magnuson said:
There was one thing very curious about the Sigma D9 specs:
it only supports Foveon RAW format - no JPEG. Now we have a news
article about native support for RAW files that states it takes
12s on a P4 to open a Foveon RAW image.

If 2+2=5, I might guess that the lack of JPEG is because it takes
too much CPU power to process the Foveon RAW image data for the
in-camera CPU to handle at any reasonable frame rate.
--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
Click to expand...
Click to expand...
 
That is before compression Erik. It is hard to guess at what it
will be actually. I got 27 by going to three times the open file
size of the G1. This could be very wrong. If it is only twice the
size of the G1 RAW, the open file will be 18 to 19 megs. I don't
know how much compression they will use, or how effective it
will be in the 16 bit format compared to the 8 bit format.
Erik Magnuson said:
Bob Williams said:
Erik, nice research, but it isn't accurate.
Click to expand...
It still does not explain the fuzzy math on why you think the RAW
file size is going to be 27MB. I'm assuming they do all they can
to reduce the file size (w/o losing data.) You seem to think that
they will attempt to maximize it. Most people think smaller files
are a good thing (again assuming no loss.)
--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
 
This is a perfect time for Sigma to use JPEG2000. Much less losy process for a given compression.
Bob the Printer said:
Bob the Printer said:
Well, lets look at the competition:
D100: RAW + JPEG
D1x: RAW + JPEG
D60: RAW + JPEG
1D: RAW + JPEG (heck the 1d can even output BOTH at the same time!)
S2: RAW + JPEG
Contax N1: RAW + JPEG

Hmm, Nikon, Canon, Fuji and Contax are all nuts! All of these
cameras record 12bits deep in RAW as well. In fact, JPEG is LESS
important on these cameras since the RAW format is smaller due to
the infamous Bayer pattern. For conventional CCDs, JPEG is so cheap
and so fast, that even the cheapest cameras have it. So again, why
not the SD9?
Click to expand...
Apparently you missed the whole point of Bob's message... The Sigma
has much more color pixel data to process than any of the cameras
that you mentioned. It would be a shame to lose this data in an 8
bit jpeg file.

But since the production SD-9 isn't out yet, we're all a bit
premature in judging it or whether or not it will have jpeg output
capabilities..

--
Sincerely,
Bob the Printer
Click to expand...
 
I find if very doubtful that computing the real r,g,b values, given the 3 sensor values, can be harder, (computationally), than doing a bayer interpolation, especially multiple nearest neighbor types.

I mean, it is just 3 equations to run, to map sensor values to rgb space. That has to be quicker than that has to be quicker than running a couple of splines, or whatever bayer does.

Is there a fixed, "optimum", way to do bayer unscrambling? Where is some info?

Thanks.

Don
Peter G said:
How could they proclaim one of their advantages being less
processing needed, if they had to icorporte even more processing
than the competition. It seems they have an expensive algorithm of
their own and they just removed the option of processing it from
the camera. Which is lame IMO.

Peter
Erik Magnuson said:
There was one thing very curious about the Sigma D9 specs:
it only supports Foveon RAW format - no JPEG. Now we have a news
article about native support for RAW files that states it takes
12s on a P4 to open a Foveon RAW image.

If 2+2=5, I might guess that the lack of JPEG is because it takes
too much CPU power to process the Foveon RAW image data for the
in-camera CPU to handle at any reasonable frame rate.
--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
Click to expand...
--Don Erway http://www.pbase.com/derway/kona_underwater_g2
 
KenEis said:
This is a perfect time for Sigma to use JPEG2000. Much less losy
process for a given compression.
Click to expand...
Good point.

As to only RAW, I would say that the SD9 will give up JPEG, but, remember this...

ALL of out CCD's interpolate. That is, the Bayer pattern is interpolated. Now, perfection would be 4 CCD pixels = 1 real life pixel, but we'd never settle for that. So, we upsample and we find that data is revealed up to a certian point, beyond which we're just wasting time. On a typical Bayer pattern, this is usually CCD pixels = screen pixels. But on Fuji's CCD, they upsampled 2x more, eg. 3mp CCD -> 6mp screen pixels, because the innovative CCD design yielded better results than Bayer. So, the Foveon X3 being a step further, which could potentially yield 3x the quality of a Bayer (ie. instead of 1 perfect pixel being an RG/GB set of 4 pixels, we collect light over all 4 pixels described as one, but 3 deep, so 3x the light gathering). Therefore, if we generate our final picture from a 3mp Foveon, we'd have 9mp. Now 9mp JPEG'ed vs a 3mp RAW is more touch and go which is faster to process and write.

So in conclusion, I think Sigma will give us JPEG, but we'd be silly not to use RAW for best quality. Bring on those 2Gb CF cards quick! (and I hope they are quick in writing!)

--Excal
 
Valliesto Bailey said:
That is an interesting point, the cost of a JPEG compressor must
have been offset by the lack of a Bayer processor. It must be much
more difficult to interpolate the color data than to apply JPEG
compression to a full color image. Seems something is amiss.
Click to expand...
Indeed, and I suspect it is this. The "development back" that we saw a picture of posted by Phil, is very very basic. It's designed for CCD testing, so it's probably got an interchangable CCD area and just points and shoots. Apature... that'll be on the lens we attach. Focus, see the lens again. Shutter speed, yeah, that'll be there. You get the idea. As to JPEG - well if you're testing a CCD, JPEG would be no good at all. And that I suspect is where the rumour about the SD9 came from... Chinese whispers again.
--Excal
 
Excal said:
And that I suspect is where the
rumour about the SD9 came from... Chinese whispers again.
Click to expand...
The lack of JPEG support on the SD9 is not a rumour. It's part of the published specs:

http://www.sigma-photo.com/Html/news/news_sd9_fs.htm

"With the RAW data recording system of the SIGMA SD9. It is possible to obtain high picture definition and compact file size. Losses compression system of RAW data, eliminates image deterioration, provides superior pictures without sacrificing original image quality, RAW data is convertible into high TIFF and JPEG formats by dedicated image processing software "SIGMA Photo Pro". "--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom
 
Erik is correct. My only wish is they change their spec again.

Here are a list of devices I can think of that do not readily support RAW - most are not that important for serious photography but adds a lot to the fun and convenience that help digital photography's success

1. Print picture directly from CF - very useful when you are running out of time. Many print shops or higher end photo printer are supporting that.
Sigma users need to go home or bring a notebook computer to the same thing?

2. Digital Picture frame - this usally give 3 times better resolution and size than the LCD screen of a digital camera. Sometimes could be useful for even for serious job (to impress the customers or friends instantly) in additional to a lot of fun and excitement.

3. PDA that support jpg - not really useful but could be fun. Since it is just for fun, nobody will do that if it is not convenient.

4. somebody's (client's or firiend's) notebook computer(s). Until Microsoft support it, other people cannot see it immediately with Sigma software. What about Apple? On several occasions, my friend used his notebook computer as backup and reviewed my pictures while I was shooting or review my pictures of another CF in my camera.

5. Future digital cellular phone that support picture. This definitely could be fun to send and receive a picture on the phone. it may only support smaller format like SD though.

6. Nixvue Vista with LCD - another example just announced at PMA

7. I believe somebody (Microsoft?) made a device that displays CF jpg on TV.

With the popularity of jpg and digital photograpy this list must be longer as time go by.

Wonder if Sigma's engineers were working so hard that they forgot the fun part of photography.
Erik Magnuson said:
Excal said:
And that I suspect is where the
rumour about the SD9 came from... Chinese whispers again.
Click to expand...
The lack of JPEG support on the SD9 is not a rumour. It's part of
the published specs:

http://www.sigma-photo.com/Html/news/news_sd9_fs.htm

"With the RAW data recording system of the SIGMA SD9. It is
possible to obtain high picture definition and compact file size.
Losses compression system of RAW data, eliminates image
deterioration, provides superior pictures without sacrificing
original image quality, RAW data is convertible into high TIFF and
JPEG formats by dedicated image processing software "SIGMA Photo
Pro". "
--
Erik
Free Windows JPEG comment editor
http://home.cfl.rr.com/maderik/edjpgcom
Click to expand...
 
You must log in or register to reply here.

Keyboard shortcuts

f
Forum
m
My threads
Mobile site
About
Editorial content
Cameras & Lenses
Community
All content, design, and layout are Copyright © 1998–2025 Digital Photography Review All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.
When you use DPReview links to buy products, the site may earn a commission.
©GPS Media - Guides, Products, Services.
Back
Top