DVCPro HD good for Aquisistion, But not Post?

[Radek Svoboda]

All these are 50 Mbps at 30p

DVCPROHD records 960x720 pixels, DVCPRO50 PAL 720x576, NTSC 720x480


By comparison with other progressive cameras:

JVC HDV records 1280x720 pixels
Sony CineAlta records 1440x1080 pixels

[Steven White]

4x, by my math

Actually, there was an error in my calculations... here's what I get:

@ 30p
Codec | Bitrate* (Mbps) | Resolution | Colourspace | Bits** per pixel | Bits per unit colour information
DV-25 | 25 | 720x480 | 4:1:1 | 2.4 | 1.61
DVCPRO-50 | 50 | 720x480 | 4:2:2 | 4.8 | 2.41
DVCPRO-HD 720 | 50 | 960x720 | 4:2:2 | 2.4 | 1.21
DVCPRO-HD 1080 | 100 | 1280x1080 | 4:2:2 | 2.4 | 1.21
HDV1 | 25 | 1280x720 | 4:2:0 | 0.9 | 0.36
HDV2 | 25 | 1440x1080 | 4:2:0 | 0.5 | 0.6

The 2nd to last number represents the bits per pixel in a single frame. You will notice, that each miniDV pixel contains the same amount of data as a DVCPRO-HD pixel (2.4 bits), and each DVCPRO-50 pixel contains twice as much data (4.8 bits), as I mentioned last night.

The final number represents the amount of data per unit of colour information per pixel, with all pixels treated equally (that is, absolutely no accounting for spatial encoding). In other words, this number is an estimate for how much work spatial encoding must do in order to make all things equal between CODECs. My conclusion therefore is that DVCPRO-HD will have more spatial artifacts than miniDV... and many more spatial artifacts than DVCPRO-50, under the assumption that all DV codecs handle spatial information similarly.

It is calculated as:
For image resolution XxY (i.e., 720x480, X = 720, Y = 480)
and colour space (a:b:c)

Bits per unit colour information per pixel = (datarate)/[(framerate)XY(a/4 + b/4 + c/4)]

Panasonic people will be happy to note how thoroughly DVCPRO-HD spanks HDV in these calculations, as obviously spatial encoding has to account for a lot of the missing information. What's not evident is how efficient the spatial encoding schemes are, nor how they balance these deficiencies to get good looking video.

The depressing result to me though is that the number of bits devoted to each DVCPRO-HD pixel are the same as the number of bits devoted to each miniDV pixel. On a per-pixel basis, I would therefore expect DVCPRO-HD to perform as well as miniDV for post work and keying. Downsampled to SD resolution, DVCPRO-HD 1080p will offer and better keys than DVCPRO-50, but the 720p format seems like it will really provide negligible improvement.

*I have assumed that all of the bitrate goes towards video in this calculation, saying nothing of how much data goes to sound.

**Note - I used "M" as equivalent to a 1,000,000. If this is not the case, and M = 1024x1024 or some other bizarre quantity, the ratio of the numbers will remain constant, but differ by a multiplicative factor.

-Steve

[Brian Sveum]

Originally Posted by Steven White

DVCPRO-HD 720 | 50 | 960x720 | 4:2:2 | 2.4 | 1.21

This is still the same as the Varicam right? That seems to have an amazing picture.

[Barry Green]

Originally Posted by Brian Sveum

This is still the same as the Varicam right? That seems to have an amazing picture.

That is what the VariCam records, yes.

As a visual demonstration, this excellent little comparison was put up on the Cow.

http://www.eleventy.org/dv-en/

It shows PAL DVCPRO25 4:1:1, vs. PAL DV 4:2:0, vs. PAL DVCPRO50 4:2:2, all encoding the same graphic. Excellent, simple, and to the point. Shows the dramatically cleaner color resolution, and much less mosquito noise, of the DVCPRO50 format.

Hopefully someone with an editing system with DVCPRO-HD on it can extend the example to show how DVCPRO-HD would compare in the same circumstances.

[R Geoff Baker]

That is a straight forward demonstration ... and the results are confusing. I'm startled to see 'vertical' artifacts in the 4:1:1 chroma resolution -- I'd have anticipated horizontal artifacts, but never the vertical ones illustrated. Even 4:2:0 would seem more likely to reveal vertical issues ...

Very curious.

GB

[Steven White]

I'd be interested to see HDV thrown into a mix like that too Barry. It was certainly a nice demonstration of the effects of compression on the images. It's an easy test to do isn't it? I would think anyone with Final Cut Pro HD could pull it off.

Alas, I've get Premiere.

-Steve

[Thomas Smet]

I think those results for 4:2:0 and 4:1:1 are backwards. In the image clearly on the 4:2:0 the blocks are 4 pixels wide and they shouldn't be. I think the images got flipped around by accident. It does still show the quality levels however.

There is a site where somebody tested most of the codecs for the Apple even DVCpro HD. I will look for it and post it later.

[R Geoff Baker]

I suspect you are right -- the images are likely mislabelled.

Unfortunately, this undercuts the confidence one can have in the results as presented.

GB

[Thomas Smet]

http://www.onerivermedia.com/codecs/index.htm

Here is the most advanced codec testing done for the Mac that you will ever find on the net.

One note however is that hopefully Apple fixed the DVCproHD codecs. I think this list is a little old now so maybe they did. Regardless of the luma bug in the codec you can see what DVCproHD will look like. I'm not sure if the Avid codec ever had this same bug or not.

I use this list a lot to show people just how bad DV really is.

Hope it helps.

[Steven White]

Hm. Well those tests certainly make all the codecs look nasty!! The one thing that troubles me most about them though, is that they're all interpolated back up to full-spec resolution... which makes me question if PA ratios have been adequately accounted for - particularly on multi-generation tests. The 1.5x anamorphic stretch on the DVCPRO-HD stuff really rips those horizontal patterns to pieces.

At any rate, this information confirms my suspician that DVCPRO-HD per pixel isn't an improvement on 4:1:1 NTSC DV. While the color artifacts appear slightly less signficant on the DVCPRO-HD, they're made up for by much more extensive spatial artifacting (mostquito noise, etc.). Kind of a shame really.

In my opinion, at 100 Mbps, I-frame compression isn't something Panasonic should be bragging about as a good thing. The images are still highly compressed, and the efficiency of adding a GOP at such data rates could do wonders for the image quality.

-Steve

[Jeff Kilgroe]

Originally Posted by Steven White

Hm. Well those tests certainly make all the codecs look nasty!!

That's for sure. I think those tests are pretty worthless though considering all the factors and what appears to be some real errors in their results. As for the 1.5X anamorphic encoding of DVCProHD, I'm a bit concerned about how that will come into play with the optics of the HVX (this isn't a $6K+ lens that we might find on a Varicam here), not to mention the smaller CCD size. I think we can expect some pretty soft HD images from the HVX, especially in 1080p.

Originally Posted by Steven White

At any rate, this information confirms my suspician that DVCPRO-HD per pixel isn't an improvement on 4:1:1 NTSC DV. While the color artifacts appear slightly less signficant on the DVCPRO-HD, they're made up for by much more extensive spatial artifacting (mostquito noise, etc.). Kind of a shame really.

Exactly. :-( We do get 4:2:2 color, but with the trade-off of higher compression. However, this is still the same format used by the Varicam and I'm sure that the HVX200 will be a huge leap forward in quality for this price range. However, I wonder how DVCProHD will hold up against the upcoming HDV50 format that has been announced and the HDV100 format that has been hinted about. I'm going to assume that the HVX200 will beat any HDV50 camera to market and it uses an established codec, so that says a lot right there. But I think before we all get too pessimistic or overly excited about this camera, we need to see some sample video clips.

Originally Posted by Steven White

In my opinion, at 100 Mbps, I-frame compression isn't something Panasonic should be bragging about as a good thing. The images are still highly compressed, and the efficiency of adding a GOP at such data rates could do wonders for the image quality.

Yes, but adding a GOP still would increase motion artifacting, even with very short GOP lengths. Ultimately, what would be the best of both worlds would be a variable GOP, variable bitrate codec. With a solid state recording media like P2, there's no technical reason to lock the camera to a constant bitrate while recording. However, I think the level of processing power needed to dynamically cope with such a codec would push this camera out of the intended price range. I'm sure Panasonic wasn't trying to reinvent everything here either as they chose to use their existing formats. DVCProHD is a proven format. I can see that for still imagery (in some situations) that ordinary HDV could give a superior image, but overall I think DVCProHD will be the superior format for most uses. But hey, if it isn't all that great, I'll save my money and just start renting bigger and better cameras to do my work instead of trying to own one.

[Tommy James]

So I am wondering how you can distribute video at a 100 megabit per second streaming rate ? I am assuming that Blu-Ray DVD can handle it. How much horsepower does your computer need? With 720p30 streaming at 20 megabits per secound my high definition plays just fine with my modest 2.5 gigahertz pentium 4 and my 128 megabyte video card. I can't play 1080p with my system.

[Jeff Kilgroe]

Originally Posted by Tommy James

So I am wondering how you can distribute video at a 100 megabit per second streaming rate ? I am assuming that Blu-Ray DVD can handle it. How much horsepower does your computer need? With 720p30 streaming at 20 megabits per secound my high definition plays just fine with my modest 2.5 gigahertz pentium 4 and my 128 megabyte video card. I can't play 1080p with my system.

Grrr... I had to edit my reply because I misunderstood what you were saying. I guess I'm tired and it's way past my bed time.

Anyway, distribution is still up in the air. BluRay and HD-DVD will probably be the media of choice, but both support MPEG2, MPEG4 and WM9 so recompression to one of those will be necessary out of the DVCProHD format. You could always place the DVCProHD file onto the BluRay or HD-DVD disc as if it were a data disc and then your intended viewer could play the file if they had the proper software or codec installed on their system. Other ways to distribute DVCPro data will probably be the same as it is now. Send it out on DVCPro tape (buy, rent, borrow, steal a deck) or give someone an HDD or DLT or DVD-R or whatnot. We can also share P2 cards, but with the current estimated P2 prices, you can have my P2 when you pry it from my cold dead fingers.

As for your computer, you have a bottleneck or problem software somewhere if you can't play back 1080p. However, you haven't given us any real specific information so it's hard to say for sure. What type of file are you trying to play? Do you have a display with full 1080p resolution or is your system/software trying to scale it down on playback? What software are you trying to play the video with? Give VLC media player a try at www.videolan.org -- it's free and is very compact with low overhead. It plays a lot of stuff like HDV and higher res video clips very well, provided you have the proper codecs installed. But a 2.4GHz P4 shouldn't break a sweat trying to play a 1080p video file if it's encoded in a reasonable format.