AVC-Ultra - revamped

I think we are talking past each other a bit...here's the crux of what I'm saying.

H.264 (or DV, or any other codec) could be used to compress RAW image data into a compressed RAW format if a version of H.264 existed which compressed the sensor data without applying gamma and color spaces. To the best of my knowledge, no such codec exists, and any and all H.264 codecs (and other video codecs) that do exist require the color data to be present at a fundamental level prior to compression.

Fact of the matter is, the sooner more camera manufacturers are happy to step up and add visually lossless compressed RAW onto media, the happier I will be. So far, it's a pretty empty field for some reason, despite it being something that many people seem to want.
 
Last edited:
The problem with the codecs we have been working with is not so much that they have color or gamma already applied. You can change the color to your heart's content. You can also change the brightness and contrast (gamma) but that's where the problem arises. And it's not because you started with a gamma but because the codecs you're using are 8 bit.

Even raw has color (in a sense) and a gamma, that you start from. But it comes to you in 10 or 12 bits.

Pop quiz: 10-bit is ________ better than 8-bit:
(A) 20%
(B) 50%
(C) 2 times
(D) 4 times

(Answer: D)

10-bit is exponentially better. It gives you 2 to the 10th power (1,024) discrete levels of brightness per color instead of 2 to the 8th (256). 8-bit color gives you 16.8 million colors (256 R x 256 G x 256 B). 10-bit gives you over 1 billion (1024 x 1024 x 1024) colors.

The problem you run into when you try to do film-like S curves in 8-bit is banding. With 10 bit, especially logarithmic 10-bit, you can do some serious stretching without banding.

--- http://www.normankoren.com/digital_tonality.html#Human_vision
 
Last edited:
Yes, with normal codecs you can change the gamma and colors, after they have been written into the file in-camera. But with RAW, that initial choice is done off-camera, with much greater control. And yes, bit-rate helps.
 
Michael Olsen said:
I think we are talking past each other a bit...here's the crux of what I'm saying.

H.264 (or DV, or any other codec) could be used to compress RAW image data into a compressed RAW format if a version of H.264 existed which compressed the sensor data without applying gamma and color spaces. To the best of my knowledge, no such codec exists, and any and all H.264 codecs (and other video codecs) that do exist require the color data to be present at a fundamental level prior to compression.

Fact of the matter is, the sooner more camera manufacturers are happy to step up and add visually lossless compressed RAW onto media, the happier I will be. So far, it's a pretty empty field for some reason, despite it being something that many people seem to want.
Click to expand...

Fair enough...and I absolutely agree that we need more good compression for RAW. Also higher bit depths...Red's using 16bit now. I know this takes a lot of processing power, but I would guess that's probably the fastest improving part of camera tech!
 
Or just completely uncompressed RAW. It's totally doable with commodity SSDs and even some CF and SDXC cards at 2K, 24 FPS. In a couple years (by the time they listen to us) CF and SDXC could probably handle 60 or more frames.
 
combatentropy said:
Or just completely uncompressed RAW. It's totally doable with commodity SSDs and even some CF and SDXC cards at 2K, 24 FPS. In a couple years (by the time they listen to us) CF and SDXC could probably handle 60 or more FPS
Click to expand...

I don't know, that still seems really wasteful when lossless gives you exactly the same thing but two and half times smaller.
 
I know it's been stated elsewhere, but even at 2K, SSD and more particularly CF/SDXC still have some speed humps to get over before widespread usability is available in off-the shelf products. Advertised rates and tested "average continuous" rates on computers for reading and writing are different than the mission critical constant data-rates required for production. A drop in data rate below a certain minimum just once a every five seconds could mean you end up missing hundreds of frames. Very few SSD/SD/CF cards/drives can manage writing absolutely constantly at ~150 MB/sec. And that also means a 2 minute shot is a bit more than 18GB. So get ready to buy a lot of the largest, fastest (two words for most expensive) media you can find.

There's a reason that RAW media - even from small manufacturers like Ikonoscop - is expensive. It has to put up with a lot of physical abuse (which many people have had problems with regarding standard SSD I/O) and they have to record absolutely flawlessly at speeds which are, right now, cutting edge.

EDIT: I just realized this has gotten very off-topic. Sorry! Hopefully we'll see Panasonic launch a compressed RAW codec for their cinema audience soon...
 
Last edited:
Don't worry, I'm slowly bringing this back around topic, too.

The nice thing about uncompressed is it takes less on-camera processing. You just write the bits to disk. With compressed you need a specialized chip. You can do it with Intels, but then your camera is big, boxy, and has fans.

Specialized chips are no problem for Panasonic. I was just thinking of the little guys. If you're Ikonoskop, uncompressed is actually cheaper and easier than compressed!

But let's say you decide to go compressed. You can either go the mainstream route (MPEG-4) or wavelets. There is one advantage if you go mainstream: specialized chips are everywhere: camcorders, DSLRs, phones, CCTV, are all buying MPEG-4 encoder chips. You could probably buy them for $20 each (in volume).

For me, I'd be fine with 10-bit 4:2:2, like the proposed AVC Long-GOP. I really think I can get the same result as wavelet-RAW. The only thing missing is it needs to go up to 60p (the literature mentions up to 30p). People say all the time that we always want more, more, more, but seriously, for me at least, we're really close to my dream cam.
 
Last edited:
combatentropy said:
For me, I'd be fine with 10-bit 4:2:2, like the proposed AVC Long-GOP. I really think I can get the same result as wavelet-RAW. The only thing missing is it needs to go up to 60p (the literature mentions up to 30p). People say all the time that we always want more, more, more, but seriously, for me at least, we're really close to my dream cam.
Click to expand...
I don't think it's a matter of whining for "more, more, more." AVCHD, a consumer codec, now has 60p. Higher-level codecs should at least have feature parity, including higher frame rates. Since Panasonic effectively controls AVC-Utra, it should bump the top AVC-LongG's top data rate to say 70-80 Mbps and add 50p and 60p.
 
You must log in or register to reply here.
Back
Top