RED RAY: claims HQ 10Mbps 4K 16:9

[Comatose]

http://www.reduser.net/forum/showthread.php?t=29369

What do you guys think? (no screens)

[poisondeathray]

What is their definition of "HQ" ? :)

[Dark Shikari]

4K is a great realm of insane marketing claims, mostly because since practically nobody has a 4K screen and because most of the "detail" in 4K footage is noise, so it's not too hard to just destroy all the fine grain and claim "high quality" compression.

[unix_sansei]

4K is a great realm of insane marketing claims, mostly because since practically nobody has a 4K screen and because most of the "detail" in 4K footage is noise, so it's not too hard to just destroy all the fine grain and claim "high quality" compression.

are you saying old JJ, Jim Jannard, is insane? ha ha

[Comatose]

Ah, and they use wavelet compression. IIRC Somebody known said a while ago that Dirac is doomed for this reason.

[Astrophizz]

I'm planning to go see Earth in 4K at a local theater this weekend. I expect my eyes to melt :)

[Manao]

The linked thread is filled by fanboys and RED RAY developers, so it's hard to get an unbiased idea. But my bullshit detectors kept ringing for the whole duration of the thread. They are also reluctant to post screenshots and example streams, so basically we have to believe that their demo wasn't faked - which hardly happens.

[*.mp4 guy]

Well.... Its a wavelet codec, wavelets deal with low bitrate scenarios very well, as they can essentially just downscale the image, so, If the demo is real, it is probably of something which contains no actual data above ~1080P ish resolution, but is sored in "4k" very effectively by wavelet compression. 10Mbit for high quality 1080P is definately doable, though still a very bold claim. I imagine in any case, the source was chosen very carefully.

[easy2Bcheesy]

David Newman at CineForm has often stated that the difference between 2K and 4K video isn't anywhere near as pronounced as the resolution leap would have you believe. I can see Red-Ray gaining favour as a standard box for playing native RedCam footage at a Digital Cinema.

[benwaggoner]

Also, is that 4K 4:4:4, or a 4K Beyer pattern (so just one sample per pixel, like RAW)? The latter is obviously easier to compress.

And as a rule, I've found believing David Newman on all things a high-success strategy for many years :). In this case, he means that there's simply less frequency data the bigger the image. You're up against the limits of the lens, the CCD (if not Beyer) etcetera.

[SeeMoreDigital]

In this case, he means that there's simply less frequency data the bigger the image. You're up against the limits of the lens, the CCD (if not Beyer) etcetera.The RED camera's incorporate their own design of capture sensors

[Sagittaire]

10 Mbps for 4K mean 4 Mbps for 1080p with the same local quality by pixel. H264 don't produce bad quality at 4 Mbps for 1080p in general case.

[vucloutr]

H264 produces nothing, maybe x264 doesn't produce bad quality at 4Mbps for 1080p in a general case...
but this would lead to RED having an encoder as good as x264. Could this be possible? I'm sceptical.

ps: and wasn't that stuff 10bit ?

[Blue_MiSfit]

0_o...

Nah...... that's too good to be true. REDCODE RAW isn't even that impressive :P

Even with the most efficient video codec in the world (x264) - you don't have much chance of making transparent (true) 4k at 10mbps. I mean, I could see it looking good, but not transparent to an uncompressed original.

They simply must have downscaled and upscaled somewhere in the post-production chain.

~MiSfit

[SeeMoreDigital]

They simply must have downscaled and upscaled somewhere in the post-production chain.If they captured some video using this camera with 4520x2540 chip (http://www.red.com/cameras/technology/) then yep, some downscaling would be in the chain ;)

[*.mp4 guy]

Using wavelets for compression is essentially as good as having variable dawnscaling/upscaling *thats how wavelets compress things* (usually). If I wanted to I could trivially modify a jpeg 2k codec so it effectively achieved compression purely by downscaling the image.

[Blue_MiSfit]

Interesting.

That's actually a pretty way to think about it. If only we could dynamically adjust resolution on a per frame or GOP basis in the DCT world..

~MiSfit

[Comatose]

Using wavelets for compression is essentially as good as having variable dawnscaling/upscaling *thats how wavelets compress things* (usually). If I wanted to I could trivially modify a jpeg 2k codec so it effectively achieved compression purely by downscaling the image.
Well, don't DCT codecs work this way too? Effectively, not intentionally.

That is, if there's fine grain - in very bitrate hungry areas (such as very dark areas) it can appear to be lower resolution, and then it will begin to block if it's highly starved. In my understanding of things, if it doesn't have enough bitrate, then the resolution is effectively reduced until it reaches blocking (which is just really, really low resolution).

Am I completely off? :P

[*.mp4 guy]

dct does not affect per pixel frequency, such as interpolation, or wavelets, but per block frequency, If you eliminate certain dct coeficients, it is equivelent to a lowpass, but it is run on all of the pixels at once, instead of individually, so you get blocking and ringing far beyond what a per pixel lowpass would cause, you can get around this by using multiple overlapping dct's, but no codecs actually do this.

[Comatose]

But, am I right in my theory that bitrate starvation effectively causes loss of resolution, similar to downsizing, in the affected areas (or blocks)?

[Manao]

Bitrate starvation will results either in artifacts or loss of details (which can be somewhat assimilated to loss of resolution). Usually, the encoder's implementation chooses the tradeoff between artifacts and loss of details (for example, psy-rd in x264 can be seen as such tradeoff)

[Comatose]

Aha, thanks :P

[benwaggoner]

That's actually a pretty way to think about it. If only we could dynamically adjust resolution on a per frame or GOP basis in the DCT world.
Can't we? Just stick in a new sequence header and a Closed GOP wherever you want to change resolution.

That's basically how Smooth Streaming works. We encode as parallel bitstreams at different bitrates and frame sizes, and then concatenate them together prior to the decoder.

While they're being sourced from different bitstreams in this example, no reason you couldn't just make a file like that.

[unix_sansei]

David Newman at CineForm has often stated that the difference between 2K and 4K video isn't anywhere near as pronounced as the resolution leap would have you believe. I can see Red-Ray gaining favour as a standard box for playing native RedCam footage at a Digital Cinema.

He's never seen Baraka.

[Astrophizz]

He's never seen Baraka.

Yeah I saw Earth this past weekend in 4k and it was pretty great but I didn't have a 2k to compare it to. I'll say that I didn't really see much grain so I'm not sure the 4k is just for noise preservation.