The holy grail (Canon Raw Rumor)

Michael Olsen said:
Shoot 4K and finish 2K is a pretty decent idea - really, it's how a lot of people work on the RED. And the extra $16000 price difference isn't just for profit. Making a 4K sensor that can read out 24 or 30fps is expensive, as is putting in the hardware and software that can take a 4K image and reliably downscale to 1080p 30 times a second, every second, for a long time. If you think heat is a problem with the 7D now...yeesh!
Click to expand...

No, twice 1080p, not 4k. Twice 1080p would be just over 4 megapixels. Basically twice as many lines as they are recording right now. 4k is closer to 11 megapixels. Almost a third the amount of data.

I don't know if that works out though. I just thought that maybe if you captured twice as many lines and then down sampled to 1080p you might be able to cut down on the problems. I don't know if that would work or how beneficial it would be if it did.
 
roxics said:
No, twice 1080p, not 4k. Twice 1080p would be just over 4 megapixels. Basically twice as many lines as they are recording right now. 4k is closer to 11 megapixels. Almost a third the amount of data.

I don't know if that works out though. I just thought that maybe if you captured twice as many lines and then down sampled to 1080p you might be able to cut down on the problems. I don't know if that would work or how beneficial it would be if it did.
Click to expand...

My math may be wrong. Here's what I'm looking at:


1080p:
1920x1080p = 2.07 Megapixels
1920*1080 = 2,073,600

1080p Times Two:
3840x2160p = 8.29 Megapixels
3840*2160 = 8,294,400

RED in 4K mode (2:1 aspect, not 16:9):
4096x2048 = 8.39 Megapixels
4096*2048 = 8,388,608


It's a lot of megapixels any way you cut it.
 
Yeah but we're talking about lines right? So you really don't need to multiply both horizontal AND vertical. You just need to double the amount of lines you're capturing.

So if you started by capturing 2,073,600 pixels you would need to capture twice as many or 4,147,200 pixels.

You're scaling vertically, not diagonally.

The attachment shows an example. The black lines are the original skipped lines and the red lines are the new added lines. You end up with twice the amount.
The math goes:

8x3=24x2=48
not
8x2=16
3x2=6
16x6=96
 

Attachments

  • grid.gif
    grid.gif
    1.4 KB · Views: 0
Last edited:
Barry_Green said:
For those of us for whom it's not good enough... that's what the S35 Scarlet is for. For the few of us who will pay the ten grand for "perfect" imagery -- it's going to be a viable product, and better than an SLR in every way.
Click to expand...

I assume you mean for footage? I can't see the still capabilities being anywhere close to a high end DSLR.
 
roxics said:
Yeah but we're talking about lines right? So you really don't need to multiply both horizontal AND vertical. You just need to double the amount of lines you're capturing.

So if you started by capturing 2,073,600 pixels you would need to capture twice as many or 4,147,200 pixels.

You're scaling vertically, not diagonally.

I could be wrong.
Click to expand...

That would be recording something like 3840x1080p (3.55:1 Aspect) or 1920x2160p (1:0.88 Aspect). Either really widescreen or almost square. Either way, it wouldn't scale down to 1080p.

In order to increase the actual frame size by two and maintain the same aspect ratio, one must double both the horizontal and vertical lines.

sizes.jpg
 
Yeah think about a square that measures 4 x 4 inches. It has an area of 16 square inches. if you want a square twice as big, you want a square of 32 square inches, which means about 5.75 x 5.75 inches, not 8 x 8 inches. 8 x 8 would make it four times as big, not two.
 
Michael Olsen said:
In order to increase the actual frame size by two and maintain the same aspect ratio, one must double both the horizontal and vertical lines.
Click to expand...

not neccessarily, you could multiply the horizontal and verticle lines by any number, as long as it's the same for both sides. you could do .7 x as big, 1.2, 5.68375, whatever, as long as you multiply both sides by the same.

to do it by two you just have to increase the AREA twice, not the dimensions
 
Jester2138 said:
not neccessarily, you could multiply the horizontal and verticle lines by any number, as long as it's the same for both sides. you could do .7 x as big, 1.2, 5.68375, whatever, as long as you multiply both sides by the same.
Click to expand...

I wrote that in order to increase the size of the frame by two, each dimension must be doubled. Specifically, two - hence, the requirement to multiply by two or double the size of both dimensions.

I don't see how you can multiply something by .7 and have the result be twice as big?

If you are just saying that preserving the aspect ratio requires the same multiplier for both dimensions, then I agree. That's a basic geometric principle!
 
Line skipping and pixel binning is bad there are no 2 ways about it.

If canon make a CMOS dedicated to video, they must copy what RED are doing and make a 2.5k to 3K raw sensor that yields 1080p rgb.

Common sense says if they do this at aps-c size the photosites will be huge.

If they use their current photosite sizes, 2/3 is a closer size and more practical.

edit:
Windowing(cropping) on the CMOS is an alternative they might go for.
 
Last edited:
My thought was that the red and black pixels were going to be combined once down sampled to 1080p.
 
Michael Olsen said:
I wrote that in order to increase the size of the frame by two, each dimension must be doubled. Specifically, two - hence, the requirement to multiply by two or double the size of both dimensions.
Click to expand...

The area of the frame you described is quadrupled, not doubled. By doubling both dimensions 2x2=4 you are quadrupling. If you wanted to increase area by 2x, you need to increase each dimension by the square root of 2 which is 1.414xxx.
 
shaocaholica said:
The area of the frame you described is quadrupled, not doubled. By doubling both dimensions 2x2=4 you are quadrupling. If you wanted to increase area by 2x, you need to increase each dimension by the square root of 2 which is 1.414xxx.
Click to expand...

I'm not talking about area. I'm talking about doubling the horizontal and vertical lines - at least that's how i interpreted the original suggestion.

This has gotten necessarily confusing!
 
roxics said:
I don't believe Canon will release a Raw camera. Not even Red is raw. It's redcode.
Click to expand...

Lets not get carried away with semantics. There are certain properties common to RAW which we want but those properties don't necessarily have to be part of a pure RAW image. I would gladly sacrifice resolution but retain all the other aspects of RAW in a video codec/format. Sure you can't call it RAW but whatever. Canon has already done this for stills and it calls it sRAW. Get it, its not pure RAW its sRAW.
 
J Davis said:
its rgb then
Click to expand...

Not quite. If you read the link and quote in my post, it describes the format as a "partial debayer". Hence, its not as heavy as a full RGB image but its also not as sharp. However, at 2.5K, downsizing to 1080p will sharpen it up. But thats not that important. The important fact is that it retains the original bit depth values without gamma correction so you still have the full latitude and bit depth compared to true RAW. You just loose resolution which you couldn't have used for motion delivery anyway.

Ultimate point is that there is an efficient middle ground between full sensor RAW and what we have now that doesn't require lineskipping or partial readout or RGB bloat. I would wager that the current Canon sRAW algorithm is much faster than a full debayer and resize.
 
Last edited:
You must log in or register to reply here.
Back
Top