Math. Lossless Redcode

[Radoslav Karapetkov]

AFAIK, the current Redcode RAW is something like a visually lossless codec, and as such, it gives 28 or 36 Mb for a second of 4K\24 fps footage.

Further AFAIK, although being visually lossless, Redcode RAW discards some information for bandwidth purposes.

I wonder if a mathematically lossless Redcode RAW is possible and whether the current RED1's CPU power can provide such recording.

Mathematically lossless codex for video and audio provide no loss of information whatsoever, thus being identical to fully uncompressed recording.

Decompressing a math.-lossless source would give 100 % identical output to a 100 % uncompressed RAW recording.

Math.-lossless sources are usually within 60-80 % of the size of the uncompressed version.

And so, a 4K\24 fps math-lossless clip would require 190-260 Mb\s, in comparison with the 100 % uncompressed 323 Mb\s.

I wonder if the RED1 could incorporate such thing in the future, thus allowing maximum quality using math.-lossless recording?

Or am I daydreaming? :nerd:

 

[Andrew Antoniewicz]

Well I guess then I will ask; what is your reason for wanting the extra bits if it is a true visually lossless image? After all bandwidth and disk space is not free.

If you can see artifacts then the algorithm is not really "visually lossless", at least for your conditions or application. Compression quality is not an absolute universal value, it is highly dependant upon the specific characteristics of the
images you are working on and how the post-compression image will be viewed. Redcode36 appears to give excellent results for natural images captured using 35mm film optics, where the images are ultimately destined to be projected for viewing. Congratulations Red Team on designing an amazing compressor design for your system.

Also remember that no one will be directly viewing the sensor image downstream. Even the most direct workflow will require that the image data be demosaiced; and I am not sure but the sensor may also be doing some creative dynamic range compression which would also need to be expanded or adjusted to match the final color space. Both of these operations will inject noise and potentially result in artifacts which all get convolved with any jitter from the compression.

Anyone know if it is possible to record a single/timelapse frame with zero compression? This would not crash the bandwidth limits and would potentially allow you to play "what if" by comparing it yourself. If so great, but I cant see a feature like this being anything more than a "cool but very low priority" feature addition.

 

[Graeme Nattress]

That's MB/s not Mb/s.... How are you going to record that at decent length on something lightweight on the back of your camera?

Graeme

 

[Radoslav Karapetkov]

I'm not denying the obvious quality of redcode, and I sincerely don't know how you've achieved it.

About the math.-lossless - perhaps I'm just being greedy for perfection (again) .

Recording on RedDRIVE?

This could be an identical alternative for full uncompressed 4K recoding - for those who want it.

From what I'm seeing Redcode36 is more than good enough.

I'm not competent on codex and how they work internally, but I'm enjoying the quality that Monkey's Audio or Flac Audio are offering, for instance. I can hear the difference, especially on monitor speakers.

The same with video lossless.

But I guess RAW is a different story.

EDIT: OK, I got it . Even with if the math-lossless option was possible, the stream would be something like 190-260 megabytes per second for 4K\24fps and that's too much for the current speeds that REDDRIVES or CFs allow.

Is this corect?

 

[Deanan]

It would be about 340MB/s or 14+MB/frame.

 

[Radoslav Karapetkov]

OK, thanks for the clarification.