DVX vs HVX: sd and filmlook
 

There is a rumour that the DVX looks better in SD then the HVX HD footage does after it's been downconverted to SD. Is this true? I've recently seen Sony Z1 footage downconverted and, IMO, the old VX/PD cameras look better in SD. Has anybody downrezzed original HVX footage to test this?

At a Panasonic show a rep was asked if the HVX will have the same filmlook (colours, gamma and latitude) as the DVX. The rep said it would look different. A Pana rep also stated that the reason the DVX looks more film-like than a XL2 is because the Canon is too sharp. I realize the HVX is capable of better colour, but with all those pixels it might be a darker and more contrasty image. How does the HVX image differ from the DVXs' image. I've heard the HVX footage described as videoy. Which camcorder produces the most film like image as far as colour matrix, gamma, latitude (and everything else) is concerned.

Thanks a lot.

my two cents
 

I shot my last two flicks on a DVX and was able to achieve a great film look with careful lighting, composition, camera movement and color correction. The problem is, if doing a film-out, the grain is too significant with such a low res capture. The HVX would perform much better in terms of a film-out since the higher resolution equates to a tighter grain structure.

Having said that, I think the most important thing in filmmaking is story, performance, technique and lastly, medium. I saw "Pieces of April" and loved it. It looked like crap but I was totally into the story. I do, however, feel that production values and how one moves the camera are incredibly important. I once saw a film shot on 35mm that looked plain aweful... almost like video. You can do a lot with any medium if you have the right skill set. This doesn't really answer your question though so I'll just say that the HVX will heavily outperform the DVX if one plans on doing a film-out. If downrezzing to SD, I see no clear advantage in getting a "film look" with the HVX.

Not necessarily. If you said "downconverted to DV" then I'd think there's more credence to the argument, but "SD" could imply DV50 or uncompressed SD; in those cases the HVX would hold a significant edge. But if converting from HD down to DV, that's where native DV origination will hold its own the best.

The thing you have to remember about the HVX is that it *is* a DVX as well. If you put it in "tape" mode, it has the exact same menu structures, the exact same options, the exact same gamma curves and image processing; the only major differences (that I can think of) are that the HVX has a sharper lens, a native 16:9 chip, and a denser pixel-count CCD. So I'll have to shoot some side-by-sides, but I think it's possible that the HVX may be a better DVX than the DVX is.

Then you have to throw DV50 into the mix. For standard-def origination, DV50 is going to offer two huge advantages: twice as much color resolution, and dramatically less compression. So for strictly standard-def origination, the DV50 mode should (emphasis on "should", as "should" and "does" don't always meet in reality!) produce a noticeably better SD picture.

Well, that's another thing about downrezzing -- HD is recorded in a different colorspace than SD is.
That's certainly the case with the Sony. However, the HVX appears, under initial reactions, to be delivering comparable latitude to the DVX series. I measured the DVX at around 7 stops, and Panasonic engineers at DV Expo said that they'd gotten "at least" 7 stops out of the HVX. Which is amazing, and I wouldn't have expected it, but looking at the footage it does seem like a valid claim.

Well, again, you've got to understand that the cameras are extraordinarily paintable. You can get nearly any look you want out of them. I've probably overused this example, but heck, let's point to it again:
http://www.fiftv.com/DVXvsJVC/TwelveMPLooks.JPG

On that page you'll see 12 different looks. Which one is the DVX? The answer is, of course, that they're all the DVX. All those shots were done in-camera, at the same location, all I did was change the menu settings. I was attempting to recreate a dozen of the looks of Magic Bullet Movie Looks, in-camera. So what does the DVX look like? Just about anything you want it to look like. And the HVX is the same way. It's just as paintable, and actually a little more capable because it has an additional gamma setting. Actually, substantially more paintable than what you're seeing in that picture, because I did those looks on the original DVX100, and the newer DVXs have three additional gamma settings and one more color matrix, and the HVX has all that plus yet another gama setting.

But if you shoot 1080p and then downrez it in the computer to SD, you will almost have 4:4:4 color space, is that not so?

That certainly wouldn't be the case with 16:9 footage which the VX and PD handled really poorly...

Most reports I've heard from Sony HDV camera owners is that their HDV footage yields better SD output than they've been able to get in the past from their DV cameras, and that's definitely been my experience. I would think that with the HVX200 it would be no contest, since your source material would have four times the data rate (compared to DV) with a lot more detail at both the luma and chroma level. Of course this all depends on how the downsampling is done and how good the original footage is, hence it would be possible to get poor results.

But think about it: most DV cameras don't come close to yielding the maximum theoretical quality which SD can offer, so there's definitely room for improvement from having better source material. My brother has done professional work with still and video imaging at a programming level and says the only time you might encounter problems downsampling from one video format to another is if they have similar but not quite the same resolution. That's clearly not the case with HD versus SD, so based on what he's saying there's no reason the downsampling shouldn't work -- assuming a properly designed downsampling algorithm.

I thought it was 4:2:2, but I could be wrong too, though.

Would it be 4:4:4? I'd defer to Graeme; I wouldn't say it is, but I can see an argument for it.

Let's take an example of 4:2:0 (in progressive scan, so as to not complicate things). In 4:2:0, each block of four pixels is actually all the same color. What differentiates them is their brightness, but the color is all the same. Only one color gets recorded per block of four pixels, in a 2x2 array.

So, how does that work? What happens if you're shooting an image that's, say, a pixel-accurate color chart? So that each of the four pixels is actually captured in different color -- say, a block of red/green/blue/orange pixels. What will 4:2:0 do with that? Well, it'll try to average all the colors together and will come up with a sort of brown.

So in 4:2:0 encoding, you'd have four different pixels of brightness, but all of them brown, even though none of them were brown to begin with. The four pixels would be lighter or darker brown, but all the same color of brown.

So now, downrez that to standard-def resolution. What that means in simplistic terms is that those four pixels will now all become one. The relative brightness of each will be averaged together to try to come up with one individual pixel that best expresses what the four pixels used to look like.

But what color will that pixel be? Brown. Even though none of the source pixels were.

In 4:4:4, you'd get color recorded for each and every pixel, so two components get recorded per pixel: the brightness, and the true color. In subsampled color you record brightness for each pixel, but color gets lumped together. In DV 4:1:1, four pixels in a row will all be the same color. In 4:2:2, two pixels in a row will be the same color. In 4:2:0 (progressive), it's done in grids of 2x2. In 4:2:0 (interlaced), well, not even Graeme tries to describe that process.

So would a true SD 4:4:4 system have recorded those four pixels as one brown pixel? And would other downrezzing/averaging systems yield color accurately, the way that a 4:4:4 SD system would? That's the question.

But furthermore, it also depends on what you intend to do with that downrezzed image. If you put it on a DVD, it'll get re-encoded to 4:2:0 anyway, merging colors together in a second round of "brown"-ness. If you put it out to DV, it'll merge four pixels in a row to become all the same color. So -- what you do with that downrezzed image plays a large part in what recorded resolution you have (or think you have).

Thanks guys,

I'm interested in how the camera images compare, if DVD is the end product. Sorry I wasn't more clear.

Well, for the sake of Chorma keying, would you get the same edges that you would at 4:4:4?

Interesting. Worth trying, certainly. The thing is, there's no exact resizing possible, so there'll be some pixel blending no matter what you do.

For example: DV is 720 wide, HDV is 1440 wide, so a 2:1 resize would give you potentially exact resizing and the cleanest most color-accurate resize. But on the vertical it doesn't work that way, 1080 doesn't evenly go into 480.

For PAL it's probably better; 540 goes a lot better into 576; you could do a 1/4 size resize and just ignore the 18 lines off the top & bottom. And at that point, perhaps it would approach 4:4:4.

With DVCPRO-HD it's not so easy -- you don't get a direct shrink size; in 1080 the pixel grid is 1280x1080, which doesn't directly correlate to 720x480; in 50i it's closer, it's 1440x1080 so at least on the horizontal you'd get the 2:1 shrink factor.

Resizing will give you one other benefit, which is decreased noise, as the noise gets averaged together (well, along with everything else). The question is: would highly compressed and then resized footage, in an attempt to approach 4:4:4, give you better end results than mildly-compressed native-res 4:2:2...

Well, we can talk theory all day long, or I can go out and shoot something and I can actually test it.

I'll put that on my list of things to do.

Barry, I am currently a college student studying to become a professional visual effects compositing artist. So when you run these tests, send me an e-mail, I can do some composites for your tests.

Please point us to this rumor?

I've not heard this, but it is possible the DVX wil do SD better than down-rezzed HVX.

It will be interesting, but somehow I feel the HVX will look better since it has a better lens and DVCPRO50.

Please point us to this rumor?

I've not heard this, but it is possible the DVX wil do SD better than down-rezzed HVX.

It will be interesting, but somehow I feel the HVX will look better since it has a better lens, and when shooting SD you have the option for DVCPRO50.


Sorry about the repeat post... Dang computer!, or should I say operator. :)

Not complicated... the 16:9 chips seal the deal...



ash =o)