32bit float vs 8 bit

[David Newman]

Originally Posted by Glenn Chan

2- In my opinion, 8-bit R'G'B' (no subsampling) is superior to 10-bit 4:2:2 Y'CbCr as chroma subsampling can produce visible artifacts (see the pictures in my article).

I disagree, and so does most of the industry. While chroma subsampling is an artifact, so is 8-bit contouring/banding, which I fell is a more significant artifact under normally viewing conditions. 10-bit 4:2:2, matches the human visual system better than 8-bit 4:4:4, that is why it is better performer.

[David Newman]

Originally Posted by Glenn Chan

Technically it doesn't. ;) Only the values 64 to 940 have legal luma levels. The values below 64 should not be reproducible (because it calls for negative light).

Negative light - nonesense. The full luma range is able to you, yes the extremes value are super-black and super-white they perfectly usable and highly valuable. Some lowend solutions just crop them off, yet these values are increased dynamic range of YUV over CG-RGB. Nearly all video cameras use these values. Now Vegas uses video systems RGB, so the 0-64 luma values map to 0-16 in RGB, so it is not even negative in the Vegas presentation. Vegas uses a wider gamut RGB so it can store the naturally wider gamut of YUV. So 10-bit YUV does have 4 times the luma levels as 8-bit 4:4:4 video systems RGB.

While the term luma is more correct, I was describing the number of discrete brightness levels, so accurate color system terminology is not needed.

[Chris Hurd]

Originally Posted by David Newman

...8-bit contouring/banding, which I fell is a more significant artifact under normal viewing conditions.

Fully agreed, David. Indeed the 8-bit quantization effects you've mentioned were an annoying problem for some owners of the original Canon XL1. It's ancient history now, but I documented this issue long ago (with a lot of help from Don Palomaki) at http://www.dvinfo.net/canon/articles/article10.php

[Joe Carney]

For the layman getting confused...32bit float is better than 10bit, way better, in terms of processing RGB color and everything else, even if you final output will be 8bit per channel. Until recently only programs like AE, Fusion, Nuke and other high end compositing apps supported it. (not counting Cineform).

Hopefully this will mean Vegas users can take full advantage of Cineform tools.

Awesome is an understatement for me.

[Emre Safak]

Originally Posted by Mark Holmes

What? What's not right about it? It's the same color space Adobe After Effects works in. It's not like it's impossible.

After Effects supports 16 bits/channel (so that's 48 bits/pixel). The Videoguys blurb says that Vegas supports 32-bit floating point color, as opposed to 32-bit RGB (8 bits/channel + alpha).

I'm curious what kind of color management support Vegas will have.

[Joe Carney]

Originally Posted by Emre Safak

After Effects supports 16 bits/channel (so that's 48 bits/pixel). Vegas supports 32-bit floating point color, as opposed to 32-bit RGB (8 bits/channel + alpha).

I'm curious what kind of color management support Vegas will have.

AE 7 and CS3 support 16bit integer per channel and 32bit float per channel.
You are referring to the way color has been referred to in marketing terms

aka 32bit color is actually 8bit per channel plus 8bit alpha.

I know it's confusing to a lot. If we referred to it the way salespeople do we could say 128bit floating point color. (adding the channels together).

Hope that helps

[Glenn Chan]

Whats your take on the new color processing or are you under a NDA?

I'll give you my thoughts after V8 ships.

[David Newman]

If you find natural world objects that are of green and magneta concentric circle then use 8-bit RGB, for the rest of the planet 10-bit 4:2:2 is simply better workflow, that is why it is the workhorse format of HDSDI (not 8-bit RGB.) Of course 10-bit 4:4:4 is better still, that is way we have dual-link HDSDI. Remember this forum primary concern and that film and video production, which requires color correction, when you push an 8-bit image you get less than an 8-bit image, coutouring occurs requiring noise to be injected, which hurts down-stream compression, which is most likely 4:2:2 or 4:2:0 anyway. Starting with more bits, color correct without banding so you can produce the best 8-bit sub-sampled deliverable. Even many film-out have gone through a 4:2:2 10-bit workflow. In the end you can't argue your way around what works particularly, the academic position that 8-bit RGB is better, doesn't holdup to what is needed in post production.

[David Newman]

Originally Posted by Glenn Chan

You're coming from an equally valid perspective in that you can map these values around in post.

Which was the theme of my last post. Color correction simply benefits from more luma bits than it does from more chroma spatial resolution. As most post production undergoes color correction of some type, most benefit the increased luma precision.

[Glenn Chan]

when you push an 8-bit image you get less than an 8-bit image, coutouring occurs requiring noise to be injected, which hurts down-stream compression, which is most likely 4:2:2 or 4:2:0 anyway.

You're still in a situation where you need to pick a poison.

With 8-bit you can get contouring. Ok.

With 10-bit 4:2:2, you either:
A- Get generation loss (if you do chroma filtering and reconstruction properly). To do the image processing properly, you need to convert from 4:2:2 to 4:4:4 and then back down to 4:2:2 for output.
B- Or you get bad quality and no generation loss (by using box or point resampling on your chroma).

In practice it's mostly a moot point since no useful camera records 8-bit R'G'B' (AFAIK... unless you count NoX???). Better yet to go with some flavour of higher bit depth and 4:4:4... which is what Red is doing with their Redcode RGB codec (no chroma subsampling), and what you can do with dual-link HD-SDI. Or lightly-compressed 4:4:4 (e.g. Cineform, HDCAM SR).

what is needed in post production.

In my opinion (and I hope you agree here!) is that lightly-compressed 4:4:4 is a better workflow and better quality than uncompressed 4:2:2. 4:2:2 is simply bad compression at high bitrates. If you look at JPEG, the better encoders only resort to chroma subsampling for the lower bitrates.

[Douglas Spotted Eagle]

For those following this thread, it's been split off as it's significantly deviated from the original post/topic.
That said, it's fun to read the differences of opinion between the academic side and the practical/perceptual side of the split topic.
Academically, I suck. I can manage the base levels of the argument, but I'll leave the math to experts like Glenn and David.
Practically, only a blind man can't see the differences between 10bit and 8bit, or 8 bit and images processed within a greater dynamic range.
As an example (for those of us that arent' math geniuses) I've rendered out the same file segment in SD MPEG.
The origin is unprocessed/corrected, improperly exposed HDV from a Z1, gain set to +3, shutter @1/60, aperture of f1.8.
Here is a zip file of both MPEGs. Nothing different about them whatsoever, except changing the bit depth.
I'll make the original MPEG clip available for you to play with yourself once V8 is in downloadable/demo form.

[Yi Fong Yu]

this is similar to what i asked here:
http://www.dvinfo.net/conf/showthread.php?t=101786

even if we could capture @4:4:4 uncompressed, edit in that mode, the final output can't be enjoyed by avg. joe's without powerful PC's. if you CAN capture, keep the raw footage+editing in 4:4:4 10/12-bit and OUTPUT in the same high color depths, i think the final results will surpass even today's blu-ray+HD DVD. video will finally come close to replacing FILM =P. i think color is the final frontier in these day and age for NLE editing... not HIGHER res. we're @a point that if we go with higher res. now (4k/8k) @home, we won't reap additional discernible quality differences with regard to how color is currently managed.

[David Newman]

Originally Posted by Glenn Chan

In my opinion (and I hope you agree here!) is that lightly-compressed 4:4:4 is a better workflow and better quality than uncompressed 4:2:2. 4:2:2 is simply bad compression at high bitrates. If you look at JPEG, the better encoders only resort to chroma subsampling for the lower bitrates.

That is way we make CineForm 444, it is a 12-bit per channel encoder. So yes we agree on this.

[Glenn Chan]

To clarify something...

I am not talking about 32bit versus 8-bit (or the 32-bitness of Vegas 8).

My comments were referring to
uncompressed 4:22 10-bit Y'CbCr
versus
uncompressed 4:4:4 8-bit R'G'B'

as formats. Both can be viewed as (small) compromises in quality... you either have the lower bit depth or you have the chroma subsampling (neither of which are always ideal).

This is not the same issue as processing images in 8/16/24/32/64/80/(10/12/14)-bit signed/unsigned integer/floating point/whatever values. You can take either of those two formats and process them in whatever type of values you want.

Sorry if that wasn't clear... I went off on a tangent. Not much to do with the 32-bitness in Vegas 8. It may have seemed that I was disparaging Vegas 8... this is not so!! (It will rock.) Move along folks...

[Jon McGuffin]

Not sure if I should apologize for being at least partially responsible for taking the subject of this forumn off topic. Very glad to see David, DSE, Glenn, etc chime in on this subject.

Still haven't actually figured out from any of the posts if the 32-bit "floating point" calculations on color is really the same thing as comparing 8-bit, 10-bit, and 12-bit but frankly, the subject is way over my head and If any of my previous posts mislead anybody, I apologize. I'd like to learn more. Any chance there is a good publication out there "Video Color Bit Depth for Dummies?"

I was merely trying to point out that the wording that Vegas has chosen to describe their new color capabilities didn't seem to make it very clear and I was a little skeptical of exactly what their claim meant. Is this concern justified? With that said, I have no doubt this feature is a plus...

In reviewing the images DSE posted regarind 32-bit vs 8-bit... I must be a blind man because the only differences I felt I could really see was in the way compressoin had been applied. It appeared there were slightly more artificats in the 8-bit version, but the 8-bit version almost appeared to have more detail to me. Not really sure, but spent 5-10 minutes going back and forth and the only advantages I could see in the 32-bit other than the artifacting issues being less is that the colors might be a bit more vibrant, but I'm not sure this is necessary a good thing either.

I'm looking at things such as the blue sticker that's on the tripod, the images of the pictures on the wall to the left. I'm also looking at the text on the sign on the back wall.

My fear is that this great new Sony feature of 32-bit processing is going to come at a huge penalty in performace which, to me, is something not acceptable. Vegas is a good performer now, but I think it could do better while working in HD formats such as HDV, etc.

Jon