.comment-link {margin-left:.6em;}

High Definition Video for Independent Filmmakers
A How To Guide for Digital Filmmakers
Welcome all! This is my blog to share my latest research,
thoughts, etc. on utilizing HD for independent filmmaking.

YES, I am available for consulting
Contact me at mike@hdforindies.com

All content copyright 2004-2007 Mike Curtis.

Friday, September 08, 2006

Amsterdam IBC 2006 - Red News - Redcode 4K RAW 

IBC 2006 Red News: REDCODE

THE FACTS:

Redcode RAW is Red's own codec:

-10 bit log
-full raster
-wavelet based
-4K RAW REDCODE is about 28 MB/sec @ 24fps (and yes, it is VBR)

OPINION:

At NAB it was stated that Red's compression solution would be called Redcode, and it would be up to 2K, 10 bit, full raster (no horizontal shrinking like some other formats do), wavelet based. They lied - it is 4K not 2K! They are talking about RAW not RGB as well. At 4K 24fps, it is about 28 MB/sec.

Errr...thunk! Come to complete stop and think about that -

Red is now saying that they can record 4K. ONBOARD. At 28 MB/sec using REDCODE RAW, which is totally doable on a single 2.5" SATA drive .

And that's megaBYTES, not megaBITS. So it is a 200 megabits/sec compressed codec (approximately). That is a HIGH data rate...and it is using wavelet technology, which is more efficient than the DCT based compression used in most camera codecs. (It also just gets soft when the data rate is too low, as opposed to mosquito noise and banding/blocking). It is also all I-frames - no frame to frame compression, which can potentially introduce artifacts as well as increase the horsepower required to edit it (although it does drop the data rate heavily, which IS useful). DVCPRO HD is all I-frames, HDV isn't. HDV at 1080i res is same # of PIXELS as DVCPRO HD at 50i, but the HDV is about 1/4 the file size, but requires more computing horsepwer to edit with.

At 28 megabytes a second, that is a heavy data rate, but it is within the capabilities of all recent, modern, decent 3.5" SATA drives. How good is the quality at that data rate? Good enought that Graeme Nattress, Red's codec guy, is happy with it. I know Graeme, and he's PICKY, so I take that as a VERY good sign. EDIT - I got to see a short sample of Redcode in the booth playing back on a 3840x2400 pixel monitor, and got inches away for 10 seconds and squinted at it - didn't see mosquito noise, didn't see banding....looks very, VERY clean. Of course, the ultimate test is to take it into post and do some color correction, see what happens when you lift the blacks, change the gamma, try to pull a color key or secondary correction, etc. - all in due time.

28 MB/sec is kinda high, but considering the source is about 325 MB/sec (12 bit linear), that ain't bad!

As for that data rate, let's chug some math:

28 MB/sec equals:
1.68 GB/min
100.8 GB/hr

...so that if you're shooting on what was called RedFlash, and is now called Digital Micro Magazine, which has a minimum capacity of 32 GB and an estimated price of roughly $1000, that is a bit under 20 minutes of footage at $50/min for storage (obviously you'd copy it to something less expensive after shooting).

...and if you were using one of the Digital Magazines (referred to at NAB as REDDRIVE), it has been discussed at NAB that those are in the 40 to 160 GB capacity range - so that'd be about 24ish minutes on the 40, up to about 95ish minutes on a 160 GB drive...

Quibbles and Concerns:

The catch is going to be this - if it is about 28 MB/sec at 24fps, then it is (since all I-frames) almost surely going to be about 35MB/sec for 30 fps. This data rate is still doable by 3.5" SATA drives, but is getting up there for the 2.5" drives they are talking about recording on. Doable, but getting to be a stretch. And what about the advertised 60fps? That'd be 70 MB/sec...more than a single 2.5" drive can do at ANY point in it's capacity. Is Red planning on a multi-drive tiny RAID type of recording device for onboard recording? Or will you have to work with the as-yet-no-details REDRAID in order to record high frame rates? EDIT - they are talking about external hard drive cages in larger sizes in the booth - multiple drives then? And can the onboard stuff compress that much data that fast to keep up anyway?

They've also talked about 323 MB/sec for 24fps 4K RAW - this would imply about 800 MB/sec for 4K @ 60fps...while I saw several RAID systems capable of this at NAB, they certainly weren't cheap, small, readily portable, cool, or battery powerable.

Some remaining questions:

-no new news on RGB codec
-no new news on 2K/1080/720 codec - RGB or RAW?
-for post workflow, writing back to the codec - can you write back to RAW from RGB? Or just an RGB codec?
-can you run it back to the camera for playout?
-I'm sure I'll come up with more, but I'm beat for now...

-mike
Comments:
...basically half the HDCAM SR data rate. Hopefully more efficient than the HDCAM SR codec. Since HDCAM SR is indistinguishable from uncompressed, it sounds like the quality of Redcode should be good enough for any indie.

I wonder if we could use it throughout the process, a la DNxHD? I know it's optimized for the whole Bayer thing but still, the codec sounds good for editing...

Bruce
 
SR uses MPEG4, but a variant that's basically I frame only MPEG1 expanded out to 4:4:4 RGB at 10bit.

REDCODE has a number of advantages - wavelet, not DCT, and it's designed for RAW, and designed for post manipulation.

Graeme
 
I guess I should change "Hopefully more efficient than the HDCAM SR codec" to "Definitely more efficient" then...

Bruce
 
Hmm. My math on what a 4k, 4:4:4, 10bit, uncompressed data rate is a bit different.

Even if we undercut the resolution a little (for easier math):
4000 * 2000 = 8,000,000 pixels per frame
8,000,000 * 3 * 10 = 240,000,000 bits per frame (3 channels, 10 bit samples)
240,000,000 * 60 = 14,400,000,000 bits per second at 60fps
14,400,000,000 / 8 / 1024 / 1024 = 1,716 megabytes per second.

The real rate is actually somewhere over 2 gigabytes/sec if you use actual 4k resolution.

What am I missing here?
 
Anonymous on math - it is RAW, as in the Bayer pattern of R, G, and B sensitive pixels, that have NOT been converted to RGB yet (read up on Bayer pattern sensors if that doesn't make sense). That cuts the data rate roughly in 1/3.

It is one channel, not 3, and 12 bits per channel for the uncompressed 4K.

Hope that helps you understand.

-mike
 
I long ago advocated that they needed at least 200mb/s wavelet for 4K, over at dvinfo, but my preference was 400mb/s. That 28MB/s (224mb/s) data rate is a bit low, it should be near visually lossless. 50MB/s (400mb/s) would have been better, it would have been much closer to visually lossless, and you can even find drives with high capacity and lowish $/GB at those data rates. Two, or more, drives covers you for high speeds.

As it is, I expect them to match the quality of good ENG codec at that resolution, which is in the range for film and would suit most of us, and much of the film industry. If you want the very best, for the very top end, the compromise would be those 3.5inch power hungry 50mb/s drives fro 400mb/s+.


I am basing these compression figures on Cineform RAW, that was developed after discussions on dvinfo Digital Cinema projects, and probably some private email. Current figure is upto 6:1 visually lossless I think.

The idea of Inter frame compression is cheap and easy to play with, as long as you do not go over board. We had one guy doing a codec for one of our project which was just to record the difference between successive frames (which is an old compression idea from the 80's) nothing else, no motion vectors or anything. Just adding a extra stage of, recording the difference between the decompressed intra frames, and has the potential to greatly aid compression ratios (if it can beat the compression ratio of the original intra compressed frame). It is a relatively simple idea, and no additional artifacts would be generated, as it is mere lossless difference compression.

It is good to see Red succeeding.


Wayne Morellini.
 
Wayne - keep in mind, 200 mbits is getting upper end of 2.5" SATA drives. 3.5" SATA could handle 50 MB/sec (with partitioning), but would be much bigger/heavier.

As to what is enough, it'd depend on how good the quality of their wavelets are - they decided against JPEG2000, so I'd assume they have something that works more efficiently than that well defined, well supported standard. Why go proprietary unless you can do better than the industry standard and all the benefits that come with it?

-mike
 
"28 MB/sec is kinda high" - that is 3 hours of data on one disk pack. Disk packs contain two disks which, if they use them in parallel, can easily handle higher rates too.

Compare that to the 15 minutes a 35mm reel gives you and then tell me you still think 28mb/sec is high! :)
 
YEAH yeah yeah - but keep in mind, I wrote that last September - before we knew a lot of what we know now.

: )

-mike
 
Well actually I'm still curious too with the difference between the math and the announced result.

I take again the others math :
4520x2540 = 12 OOO OOO pixel
12M x 10 bits = 120 000 000bits
120M x 24 fps = 2 880 000 000bits/s
2 880M /8/1024/1024 = 343MBytes/s

So ok I know I miss the RAW, but if tou calcule this, it seems that the RAW is a 1:11 compression !

So for me, 27MB/s seems really low. Just think that it's 216Mb/s and just the SD-SDI is 270Mb/s. How that's possible ?
 
Post a Comment


Links to this post:

Create a Link

This page is powered by Blogger. Isn't yours?

Listed on BlogShares