REDCODE is 10:1 and visually lossless?

[combatentropy]

RED team, please confirm:

4K REDCODE RAW:
- compresses at about 10 to 1
- is visually lossless

Those two facts seem contradictory.

RED advertises a data rate for 4K REDCODE RAW of 27 MB/sec.
4K uncompressed would be about 270 MB/sec.
The math:
12 bits per pixel
- times -
4000 x 2000 pixels per frame
- times -
24 frames per second
- divided by -
8 bits per byte
- divided by -
1024 bytes per kilobyte
- divided by -
1024 kilobytes per megabyte

[Nick Shaw]

Wavelet compression gives very good compression ratios with negligible quality loss.

[Nick Shaw]

Also, footage from the Red One is very clean, so compresses very efficiently. It has been suggested by Graeme that REDCODE (RGB) might not be suitable (at least at in camera data rates) for compression of non Red originated material.

[Brandon Fraley]

yes, just accept it... and celebrate :D

[Nick Shaw]

In fact, to be pedantic, since Red 4k is 4096x2304, uncompressed RAW would be 324MB/s, so 27MB/s is 12:1 compression. Data rates have not been absolutely finalised yet, but we have been assured that Red will not allow data rates to be used which fall below their (undoubtedly high) quality criteria.

[GlennChan]

You could try this out with Photoshop and JPEG export... a lot of stuff will export with high compression and be visually indistinguishable from the original (at 100% zoom). At the really high bitrates, I don't believe there is a big difference between JPEG and wavelet-based compression.

2- Redcode RAW is a little different since it compresses the signal before any image processing, and on the RAW signal (not RGB). Compressing the RAW signal is a little easier since any image processing adds noise and/or makes the image harder to compress. On top of that, the mysterium sensor doesn't have a lot of noise (which makes the image easier to compress).

The Red images also have a lot of optical low pass filtering in front of them (OLPF is essentially blurry glass). Images without a lot of high frequency detail are easier to compress. In Photoshop's JPEG export, you have a blur slider for exactly this reason.

Sensors need OLPF to avoid aliasing, and I believe Bayer needs OLPF more so. In camera systems, you generally have trade-offs between:
--sharpness/resolution
--prone to aliasing
--ringing artifacts / halos (i.e. oversharpening does this)

You can't have everything, so you have to pick your poison.

The Red camera (from what I've seen) leans towards no halos, not very prone to aliasing, and not perfectly sharp. Incidentally, not very sharp images are easier to compress. "Not perfectly sharp" may sound like a bad thing, but remember that:
-You have 4K sampling to begin with, so you're getting a lot of resolution. I believe the figure being thrown around is that bayer cameras have about 70% of the theoretical maximum resolution (though I forget the figures Graeme quoted). For 1080p or 2K release, you shouldn't have any problems.
-You can use sharpening tricks so that the perceived sharpness is fine.
http://www.cambridgeincolour.com/tutorials/sharpness.htm

Granted, you can do better. But there are few practical systems that do so.

3- See for yourself:
http://www.cinematography.net/Red/comp-matrix.html

*I believe Redcode RAW has changed / gotten better since then.

[Stephen Gentle]

Granted, you can do better. But there are few practical systems that do so.

3- See for yourself:
http://www.cinematography.net/Red/comp-matrix.html

*I believe Redcode RAW has changed / gotten better since then.

The whole camera would have got a lot better since then - that stuff was taken with Frankie, wasn't it? (Frankie was the first prototype for anyone who doesn't know)