Convert 4:2:0 video to 4:2:2 mpeg2 video in conjunction with Sony Vegas-possible?

[Graham Bernard]

My understanding is that colour space is decided at in-camera capture. I can’t see how one could up-colour space from 4:2:0 to the greater colour space of 4:2:2?

But I’m willing to learn.

Grazie

[Angelo Mike]

There are programs that let you do that, including Cineform Neoscene. But Cineform converts footage to avi, which I can’t use for this project.

http://www.scenethroughglass.com

[Jeff Schroeder]

you’re right it happens in the camera. But it can be converted afterward, what you end up with is 4:2:0 sampling in a 4:2:2 space.

Windows 7 64-bit, ASUS P6X58D, i7 960 3.20GHz, 24.0GB DDR3, 12TB connected storage

[Graham Bernard]

Does that mean that within the 4:2:n the “n” remains empty? How can this be useful? How can this be useful to sampling? If there wasn’t any material captured in-camera, it still won’t be there? Details in white clouds would still be missing, surely?

I really don’t understand this at all……

Grazie

[Mark Thompson]

I had read somewhere (of course I’m unable to remember where) that Sony Vegas does import to a better color space. However if you simply render this out it will make no difference, you have what you have from the camera. On the other hand if you color correct or grade that video then you may well be adding colors that were not there in the original.

It made sense to me that that vegas could do this

[Mark Thompson]

Having said that, I browsed render settings and couldn’t find one that matches the original requirement.

Sony MXF offers a 422 Profile, but that’s probably of no help

[Dave Haynie]

No… what it means is that the additional color samples are interpolated. Like this… here’s you 4:2:0 sampling. That means 1/2 pixel rate for sampling, both vertically and horizontally. Each S (CrCb sample) is derived from the average of four pixels on encode, and applied to four pixels on decode, in a 2×2 block… to make it simple in text, I put the “S” at the top left of the block it applies to.

1: S x S x S x S x S
2: x x x x x x x x x
3: S x S x S x S x S
4: x x x x x x x x x

In moving to 4:2:2, you expect color sampling at half the horizontal rate, but the full vertical rate… in my sketch, each sample applies to two pixels. Here’s the interpolated 4:2:2 sampling

1: S x S x S x S x S
2: s x s x s x s x s
3: S x S x S x S x S
4: s x s x s x s x s

But of course, you don’t have samples at that grain, so you have to make some. You could just fake it: set s = S for the same block. That gets you 4:2:2, but it has no more actual color resolution than 4:2:0. Or we can interpolate: s(n) = (S(n-1)+S(n+1))/2. This will create color samples from the block above and below. That’s new information, but it’s not necessarily any more accurate. In a very smooth gradation of color, this might actually guess well. But on the other hand, the S(n-1) sample is averaged from the color that used to be as s(n). The S(n+1) sample might be close, or it might hit some kind of boundary, making the interpolation actually worse than the original. Ideally, a smart algorithm can use a combination of the two, or even information from the prior and next frame, to better fill in the missing information. Not that it’s commonly done, but it’s certainly possible.

-Dave

[Angelo Mike]

Thanks, Dave. At this point I may try that. I tried a free trial of Encoder Pro, and I’m waiting to hear from the broadcaster whether the test upload I did at least has 4:2:2 color.

Uh, so how do I implement your method of faking 4:2:2? This video has become increasingly problematic-a client gave me a 720×512, PAR .9841 video that a company they worked with shot two years ago, and I’m having to render it to 4:2:2, along with moving text that they have that’s not in the safe area into the safe area for broadcast. So I’m having to fake everything, really-cover up current text and replace it with text in the safe area, and strangest of all, make this video fit into a 720×512 render.

What I mean by that-and why I’m up at 3 AM-is that on the videos I got, there’s a horizontal black space, like a bar, at the top of the video. So the video itself, without that black bar, is more like 668×400. Everything important is out of the safe area. I may have finally dealt with some of that, but getting the video to 4:2:2 will hopefully be another step closer.

The company that originally shot the video doesn’t have any digital copies of the video itself. Just DVCPro and minidv tapes of their footage, which I’m hoping not to have to go through in order to recreate the entire commercial since that would delay this project even more.

http://www.scenethroughglass.com

[Dave Haynie]

[Angelo Mike] “Uh, so how do I implement your method of faking 4:2:2?”

Simply put, if you render to a 4:2:2 format, you’re going to get a 4:2:2 video. That’s just as much a part of the video format as the resolution or compression; any NLE rendering to 4:2:2 will deliver 4:2:2; any NLE rendering to 4:4:4 will deliver 4:4:4, etc.

The only question is whether you can control the up-conversion or not. Vegas doesn’t provide any way I know of to direct how a color interpolation would be done. Most NLE work is going to be done in 4:4:4 upscaled color anyway, far as I know, just as most audio DAW work will be done in floating point or 48-bit fixed point or some other format that’s much better than the typical storage format.

Why would it matter? If the end product is a 4:2:2 output, you may benefit some from an intelligent interpolation. If the video’s going back to 4:2:0 at some point (eg, for DVD or other consumer video output), you might not want to do anything too clever. The encoder, of course, has no idea where the video came from, what kind of processing was done, etc. So it’s going to naturally bucket and average color pixels back to 4:2:0… if you got too clever on the upconversion, it might wind up looking worse. Because of this, I suspect the NLEs are going to be pretty tame about how they map color decimation into internal format.

-Dave

[Angelo Mike]

I know what you mean. But, that’s the broadcast specifications I have to meet, and the videos were completed by another company that no longer has rendered versions or files they can render to let me edit them without all the problems of matching broadcast specs I’ve been having. I may be able to see if they’ll accept 4:2:0 given just how bizarre this has all become (if you can’t tell, I’ve been posting about this video all week as well as probably researching for many hours on how to fix all of this).

I could Pro Res the video with a FCP editor I know, but the broadcast specs require Pro Res video to be 720×486, which will not work with this video since it creates pillar boxing. So I’m left in the same spot.

So is there some other program that will do it? I tried a trial of Encoder Pro, though I’m not sure if it changed the bit rate of the video, which needs to stay at 18 Mbps.

http://www.scenethroughglass.com