Raw v sRaw

[KevinA]

DO Try this at home, take a 1DsmkIII set it to iso 1600, load both card slots, set one to RAW the other to sRaw, shoot in low light, resize the sRaw to Raw size (5616 x rip) in Photoshop or similar, view on screen at 100% and tell me what you see?
The difference to me is so small it's not worth mentioning. I would like to hear others results. The biggest difference to me is the sRAW colour out of the can is much better, wht would it be different?

Kevin.

[DarkPenguin]

DO Try this at home, take a 1DsmkIII set it to iso 1600, load both card slots, set one to RAW the other to sRaw, shoot in low light, resize the sRaw to Raw size (5616 x rip) in Photoshop or similar, view on screen at 100% and tell me what you see?
The difference to me is so small it's not worth mentioning. I would like to hear others results. The biggest difference to me is the sRAW colour out of the can is much better, wht would it be different?

Kevin.

I don't have one so I can't do the exercise.  But as to the color wouldn't the camera  have to demosaic the sensor data in order to create the sRAW file?  I should think that that would bake some color information in.

[digitaldog]

I don't have one so I can't do the exercise.  But as to the color wouldn't the camera  have to demosaic the sensor data in order to create the sRAW file?  I should think that that would bake some color information in.

Thats the problem with sRaw, its not Raw (its demosaiced data), making for those of us wishing to render Raw data quite questionable in terms of any usefulness.

[KevinA]

Thats the problem with sRaw, its not Raw (its demosaiced data), making for those of us wishing to render Raw data quite questionable in terms of any usefulness.

Whatever it is I can't see any worthwhile difference in resolution, detail or noise from the RAW compared to the sRaw.

Kevin.

[KevinA]

Whatever it is I can't see any worthwhile difference in resolution, detail or noise from the RAW compared to the sRaw.

Kevin.

I would of thought the true RAW at 21mp against the sRaw at 5.2 would be head and shoulders above it detail, noise etc but it's not.

Kevin.

[digitaldog]

Whatever it is I can't see any worthwhile difference in resolution, detail or noise from the RAW compared to the sRaw.

The point is you can’t put the toothpaste back in the tube, its baked pixels not Raw. The benefits of being able to render the data from Raw is lost. Got nothing to do with resolution, detail or noise, its all about rendering control.

[KevinA]

The point is you can’t put the toothpaste back in the tube, its baked pixels not Raw. The benefits of being able to render the data from Raw is lost. Got nothing to do with resolution, detail or noise, its all about rendering control.

Not sure what a "baked pixel" is. I'm still confused as to why 5.3 mp captures resized up to a 21mp holds as much detail as the native 21mp  capture. Is it like a jpg compressed, but would it not open as 21mp size? I thought shooting at sRAW would mean the camera is dumping lots of information to bring the file size down to a quarter of the original, so noise and resolution would be the poorer for it, but that does not appear to be the result.  So what is sRAW?

Cheers,
Kevin.

[Josh-H]

Not sure what a "baked pixel" is. I'm still confused as to why 5.3 mp captures resized up to a 21mp holds as much detail as the native 21mp  capture. Is it like a jpg compressed, but would it not open as 21mp size? I thought shooting at sRAW would mean the camera is dumping lots of information to bring the file size down to a quarter of the original, so noise and resolution would be the poorer for it, but that does not appear to be the result.  So what is sRAW?

Cheers,
Kevin.

Try this - shoot the same dark scene at ISO800 on the 1DSMK3 in both RAW and SRAW and underexpose the shot by 2 stops.

Open both files and correct the underexposure - then tell me what you see at 100% (I am tipping you wont shoot SRAW after this exercise)  

A baked pixel is simply a reference to a pixel in any image form other than true RAW. In other words - its already been demosaiced.

[Slough]

My ignorant explanation is this. RAW contains data from each pixel, and as each has a colour filter in front, each pixel is in effect a colour value (R, G or B, or whatever set they choose to use). To get the 21MP image you must demosaic the RAW. For each channel where are 7MP of data. Therefore to get the full monty for each channel you need to interpolate i.e. uprex from 7MP to 21MP.

The sRAW contains data that has already been demosaiced, and then reduced to 5.3MP. So to get 21MP, you need to interpolate. Well 5.3MP interpolated to 21MP is not so different from 7MP interpolated to 21MP.

No doubt someone who knows what they are talking about will know if this is correct. But I think it sort of explains what is going on.

I am still a bit surprised at the original poster's results. You would think that 21MP, albeit with a Bayer grid in front, would contain more spatial information than a simple 7MP array of pixels sensitive to all colours. After all, don't the spectral distributions from the Bayer filters overlap significantly?

[KevinA]

Try this - shoot the same dark scene at ISO800 on the 1DSMK3 in both RAW and SRAW and underexpose the shot by 2 stops.

Open both files and correct the underexposure - then tell me what you see at 100% (I am tipping you wont shoot SRAW after this exercise)  

A baked pixel is simply a reference to a pixel in any image form other than true RAW. In other words - its already been demosaiced.

I don't intend to shoot sRAW, I just did an experiment to see what it does. I was more than a bit surprised to see little or no difference, even after a Photoshop upsize. At 100% you could see slight differences not always to the sRAW's detriment. At print size you would struggle to spot anything. So despite it's technological inferiority in practice the difference is small and if it is so inferior I think that makes it all the more remarkable.
[attachment=17350:A1.jpg][attachment=17354:A.jpg][attachment=17351:B.jpg][attach
ment=17353:C.jpg][attachment=17352:B2.jpg]

[Christopher]

I don't intend to shoot sRAW, I just did an experiment to see what it does. I was more than a bit surprised to see little or no difference, even after a Photoshop upsize. At 100% you could see slight differences not always to the sRAW's detriment. At print size you would struggle to spot anything. So despite it's technological inferiority in practice the difference is small and if it is so inferior I think that makes it all the more remarkable.
[attachment=17350:A1.jpg][attachment=17354:A.jpg][attachment=17351:B.jpg][attach
ment=17353:C.jpg][attachment=17352:B2.jpg]

Just uprezz both to a 20x30 print and print both, you will see a much larger difference. However even in your samples the difference is quite easy to see. For example the text on the blue book.

[KevinA]

My ignorant explanation is this. RAW contains data from each pixel, and as each has a colour filter in front, each pixel is in effect a colour value (R, G or B, or whatever set they choose to use). To get the 21MP image you must demosaic the RAW. For each channel where are 7MP of data. Therefore to get the full monty for each channel you need to interpolate i.e. uprex from 7MP to 21MP.

The sRAW contains data that has already been demosaiced, and then reduced to 5.3MP. So to get 21MP, you need to interpolate. Well 5.3MP interpolated to 21MP is not so different from 7MP interpolated to 21MP.

No doubt someone who knows what they are talking about will know if this is correct. But I think it sort of explains what is going on.

I am still a bit surprised at the original poster's results. You would think that 21MP, albeit with a Bayer grid in front, would contain more spatial information than a simple 7MP array of pixels sensitive to all colours. After all, don't the spectral distributions from the Bayer filters overlap significantly?

Making a bit more sense.
I have only upsized with bicubic smooth nothing fancy, it produces a bit more grain. I had a 30 x 20 made of a night shot, I could not fault it. I really thought there would be a huge difference, not a pixel peeping one.

Kevin.

[KevinA]

Just uprezz both to a 20x30 print and print both, you will see a much larger difference. However even in your samples the difference is quite easy to see. For example the text on the blue book.

In print the difference is less, if you don't have the two side by side you would not bet a penny on any of them as to which they are. I picked that crop as one where the difference looks greatest. I'm not arguing a case for shooting sRAW, I was amazed at the little difference especially in print between a 5.2mp upsized file and a 21mp native file. I have a 20 x 30 print on the desk, if anything would give it away on close inspection its slight jaggies on a couple of diagonals and you need to be on top of it to see them. Those crops are minus any sharpening or fudging around, straight out of Photoshop same minimal settings with a then bicubic added to the sRAW for same size comparison.
By all comments the fact that it is a sRAW and not a RAW should make the difference bigger. Seriously if you opened the sRAW without knowing would it hit you as second class?
Should you not expect 21mp RAW to be out of site of 5.2mp sRAW? I did.
Kevin.

[Josh-H]

However even in your samples the difference is quite easy to see

Agreed - the difference is considerable  - you can clearly read the text on the small blue book in the RAW - its mush in the SRAW.

[MatthewCromer]

Looks to me like the sRAW shot is suffering from camera shake. . .

Were those really both the same image, one written as RAW and the other as sRAW?

The difference is huge, enormous, massive in terms of fine detail -- so much so that I have to make sure they were both the same shot!

[KevinA]

Looks to me like the sRAW shot is suffering from camera shake. . .

Were those really both the same image, one written as RAW and the other as sRAW?

The difference is huge, enormous, massive in terms of fine detail -- so much so that I have to make sure they were both the same shot!

Sorry guys forgot to mention the small but probably important detail that they are crops.
[attachment=17363:_X2T3208.jpg]

Kevin.

[digitaldog]

Not sure what a "baked pixel" is.

I told you above. its not Raw (its demosaiced data). Again, its got everything with rendering the data from Raw, not resolution etc.

See:

http://wwwimages.adobe.com/www.adobe.com/p...renderprint.pdf

[KevinA]

I told you above. its not Raw (its demosaiced data). Again, its got everything with rendering the data from Raw, not resolution etc.

See:

http://wwwimages.adobe.com/www.adobe.com/p...renderprint.pdf

Thanks Andrew.

Kevin.

[KevinA]

Thanks Andrew.

Kevin.

The reason I tried sRAW was to see if it was perhaps less noisy at high iso. I thought wrongly as it turns out it would be like using a 5mp camera resolution wise, with light gathering of 21mp,  I bet most people thought the same or similar. After all the native file size of a sRAW at 300dpi is near to 10 x 8 and the file size is about one quarter of the normal RAW. I also reasoned that because it is called RAW with a small "s" in the front, it would be the same as a native RAW, it opens with the same tools in Photoshop as a normal RAW. The Canon manual suggests it's just a RAW but smaller.
So when comparing an upsized "s" tiff to a normal tiff I was a bit surprised to see little difference, in fact I had 20 x 30 prints made and the difference is negligible. The missing 16mp was a mystery to me, the difference does not equate to 16mp as I look at the prints, i thought the place it would show up big time was in resolution. Hands up those that thought the same.

Kevin.

[Panopeeper]

Kevin,

1. your samples prove, that the difference between the upresed sRaw and the full raw is huge.

2. draw a small matrix of raw pixels and envision what is happening.

a. In the full raw image, each pixel carries one component of the "final" color, The other two components have to be "created", calculated from the neighbours. This is an educated guess.

b. In the sRaw image the pixels of different colors are "integrated". You have read the term "subsampling", but that's incorrect; there is no subsampling here, because there is nothing to subsample from. Anyway, now you have tri-color pixels, but they are "farther apart".

c. When you are upresing the sRaw, the interpolation, whatever method you select, is creating new pixels; this too is guessing. However, this is a "lesser educated guess" than the demosaicing was, because

I. the demosaicing needs to guess only two of the three colors for each pixel,

II. that pixels serving as basis for the guessing are closer to each other.

Therefor you can not get the same resolution with upsampling the sRaw; but that's all right, for that is not the purpose of sRaw.