CMOS of the RedOne

[Julien Lambert]

Dear all,

My name is Julien, I'm french and I'm looking for specific info on the RedOne. "Bravo" for this website, it's great!

1. Can you explain me exactly why the RedOne has 4900x2580 pixels, than we get 4520 x 2540 activ pixels? Usually, this is for the black balance. Is it true here?
In a 4K 2:1 ratio, we have 4096 x 2048 pixels, the 424 other pixels (4520-4096) are used for the DP in order to see the out-border of the frame, right? I would like to know precisely what are those pixels used for.

2. I've read the "why cmos" post (very interesting!), but i couldn't find how does the RAW converter (i mean here after the matrix) get the information about chroma whereas we have 50% of green, and 25% of Red and blue pixel. How does it get the information of the two other colors and the luma (average of the 8 pixels above?).
http://www.julien-lambert.com/online/bayer.jpg

3. Does anyone has a good URL to introduce precisely the "RedCode Raw", to explain how it works with the wavelet compression?


Thanks a lot and sorry for my english, I've to improve it! ;)

Julien

www.julien-lambert.com

[Andrew M.]

You get 4096 unique reading for each line of the sensor kind of this way (logicaly)

http://www.reduser.net/forum/uploaded/805_1210680106.jpg


Also try this.
http://www.reduser.net/forum/showthread.php?p=1349#post1349

and this
http://www.microsoft.com/windows/windowsmedia/howto/articles/vc1techoverview.aspx

http://en.wikipedia.org/wiki/Wavelet

[Julien Lambert]

Andrew, i've just read the theads you gave me.
I guess you've asked the same question yourself. Thanks, it helps me a lot.

If someone has info about the point 1 (pixels arrangement) et 3 (about REDCODE through the wavelet compression), it would be nice.

Thanks, Julien

[Stuart English]

1. Yes to the look around question.

2. Andrew offered a pretty good explanation.

3. REDCODE does use Wavelets, but its optimized to the characteristics of the Mysterium sensor's RAW data, so we don't release details about that obviously.

[Martin Weiss]

http://www.reduser.net/forum/uploaded/805_1210680106.jpg

So it´s not just a rumor that Build 16 will include Q-bert? (link (http://www.youtube.com/watch?v=b-LR3036LuI))

[Julien Lambert]

Thanks Stuart.
Martin, nice, Q-bert in 4k obviously...! ;)

[Stuart English]

So it´s not just a rumor that Build 16 will include Q-bert?

Just working on colorizing the rear LCD display. Of course you will also need headphones.

[I Bloom]

Crystal Castles kicks Q-berts ash. We're gonna need this tool on set.

I wrote this explanation about raw conversion you might find helpful:
http://www.redhax.net/wiki/Bayer_Demosaicing

IBloom

[Axel Mertes]

Just working on colorizing the rear LCD display. Of course you will also need headphones.


Damn it, so the monchrome Tetris is going to be abandoned like the RAW port was? I was really waiting for it to be released! :bleh:
Why do all these guys around want COLOR from a sensor? :wacko:

Maybe we can at least remain with the screen saver with auto-black shading...

Cheers,
Axel

[Julien Lambert]

Thank you Stoopercraft. I knew that RedCode is close to JPEG 2000, but I would like to undertand what are the differences. if someone knows that more precisely.
Ibloom, thanks a lot, your explanation about demosaicing is very interesting and It helped me a lot to understand about the interpolation of the bayer.

Ciao. Julien

[GlennChan]

Thank you Stoopercraft. I knew that RedCode is close to JPEG 2000, but I would like to undertand what are the differences. if someone knows that more precisely.
Redcode is designed for RAW images (bayer pattern), not RGB (or grayscale) images. If Red were to use JPEG2000, debayering would have to be performed and then the resulting RGB values would get encoded. Or JPEG2000 would be applied onto the RAW image converted into a monochromatic image. That would not compress very well because non-grey areas will have lots of high frequency detail.

Or another way of thinking about it... using a RGB codec would be inefficient because you need to compress a lot of made-up information.
Using a grayscale codec would also be inefficient because of high frequency detail on non-achromatic areas.
So Redcode is optimized for RAW images to avoid those problems.

2- Redcode is 12-bit. JPEG2000 has different bit depths.

Redcode RAW compresses the RAW image without image processing. Image processing tends to add information so there is more data to compress. Or it can also throw away information.

They are also probably doing some tricks because the encoder can assume that its data came from the Mysterium sensor and it knows characteristics about that kind of image, and what the post processing is (because Red also does their own deBayering).

3- Please see Stuart's response.


3. REDCODE does use Wavelets, but its optimized to the characteristics of the Mysterium sensor's RAW data, so we don't release details about that obviously.

[GlennChan]

Julien,
To add on...

I don't think it's very helpful to theorize about Redcode. I could come up with any number of reasons why it might be good (e.g. tailored for Red's sensor and deBayer, high degree of specialization) or bad (e.g. JPEG2000 has many experts working on it).

You'd be much better off looking at the end result.

You can download some TIFFs here showing an earlier version of Redcode versus uncompressed. (The codec has improved since then.)
http://www.cinematography.net/Red/comp-matrix.html

There is also greenscreen r3d files up on redrelay.net

[Julien Lambert]

Hello GlennChan,

Thanks for posting. I know that theorizing about RedCode is not that important. I am working on a writen project about the Red and thats why i am looking for that kid of information.
Thank you!

[jbeale]

Note also, Red is ("Real Soon Now") to release a new version of Redcode with updated firmware for the camera. My understanding is that this involves a ground-up rewrite of the codec. Details have not been released on what's changing, but if your project goes into any specifics on the Redcode design, it may soon be obsolete.

[Axel Mertes]

Redcode is designed for RAW images (bayer pattern), not RGB (or grayscale) images. If Red were to use JPEG2000, debayering would have to be performed and then the resulting RGB values would get encoded. Or JPEG2000 would be applied onto the RAW image converted into a monochromatic image. That would not compress very well because non-grey areas will have lots of high frequency detail.

Or another way of thinking about it... using a RGB codec would be inefficient because you need to compress a lot of made-up information.
Using a grayscale codec would also be inefficient because of high frequency detail on non-achromatic areas.
So Redcode is optimized for RAW images to avoid those problems.

2- Redcode is 12-bit. JPEG2000 has different bit depths.

Redcode RAW compresses the RAW image without image processing. Image processing tends to add information so there is more data to compress. Or it can also throw away information.

They are also probably doing some tricks because the encoder can assume that its data came from the Mysterium sensor and it knows characteristics about that kind of image, and what the post processing is (because Red also does their own deBayering).

3- Please see Stuart's response.

Hi Glen,

JPEG 2000 can do a lot more than many end users think of. So you can store e.g. RGB+Alpha in JPEG 2000. Further it allows different bit depth, such as 8, 12 or even 16 bits per pixel. It allows to assemble large images from smaller window chunks. And so on and so on.

Lets see what Build 16 is bringing us and how the new codec will look like.

Axel

[sbaechler]

Note also, Red is ("Real Soon Now") to release a new version of Redcode with updated firmware for the camera.

Don't forget the long time ago promised Redcode encoder for Quicktime / Final Cut.

[jwwoods]

When you say Redcode is 12 bits deep, does this mean that the Mysterium CMOS sensor sends out 12 bits per pixel?

Thanks,
John

[Daniel García]

Don't forget the long time ago promised Redcode encoder for Quicktime / Final Cut.

Really? They promised that? Does anyone have a link?

[Martin Weiss]

Found a nice video from a lecture, demystifying what Bayer does.

http://media.panavision.com/ScreeningRoom/Screening_Room/Demystifying_Part6_480p.html

Enjoy.

[Graeme Nattress]

It's certainly an interesting video, but elements of bayer description are so wrong, see http://www.reduser.net/forum/showthread.php?p=226134#post226134

Graeme