As several of you pointed out in other threads on the XDCAM EX, the difference between the HVX-200 and the PMW EX1 is not simply 4:2:2 vs. 4:2:0. Please check my math and let me know if I have miscalculated.

XDCAM EX has full 1920 x 1080 CMOS sensors.

For 720P it provides full 1280 x 720 = 921600 pixels / frame (and luma samples / frame)

With 4:2:0 chroma subsampling divide by 4 = 230,400 Cb, Cr samples / frame

The HVX 200 has an actual sensor resolution of 960 x 540 = 518400 luma samples / frame ... (960 x 720 if you consider pixel shifting for luma... which doesn't help chroma resolution)

With 4:2:2 chroma subsampling we divide by 2 = 259,200 Cb, Cr samples / frame

So the PMW EX1 ends up with almost twice the luma samples and almost the same chroma per frame.

For 1080i,p the difference is even greater... 4X the luma resolution and 2X the chroma samples per frame. If you capture HD SDI at 4:2:2 you get 4X the chroma samples from the PMW EX1.

I don't have much experience chroma keying... are there other factors in the signal besides the color resolution that affect keying or applying effects?

TV

Yes, thanks Alexander, it does.

A shame that. Rather than just filling the 4:2:2 signal with the 4:2:0 source, it would be more useful to convert it to an extrapolated 4:2:2, I should think.

I am primarily concerned with progressive, rather than interlaced. Guess I need to educate myself on what impact that has on the color sampling scheme. I read the Wiki articles on color sampling mentioned in earlier posts, think I'll delve into the DTV Handbook, as their coverage is usually quite expansive.

I guess now I have to do some research on keying and compositing what are essentially 4:2:0 signals. Once captured by an AJA card as HD SDI, would any conversion be required to round trip to AE? Any resources that you could steer me towards would be appreciated.

I was hoping to convert that HD SDI signal into ProRes 422. Still have to figure out how to transfer that into AE for grading, CC, and efx.

Unfortunately, knowing the answere and having a solution are not one and the same thing, but thanks for a thoughtful answer.

I think I get what he's getting at.

Short answer: NO.

Once the signal is recorded to the SxS, its 4:2:0- that's it.

It is in fact converted to 4:2:2 for playback over the HD/SDI. This is required by the HD/SDI specification. (SMPTE 292M) Its a bit hard to get the actual spec... as the SMPTE sells it for way too much.

That conversion does not interpolate or otherwise try to display additional information. All of the "extra" information is zeroed out- i.e. it contains no information.

The signal coming out is the same data that you have on the SxS, and a lot of wasted bandwidth. Quality wise it might as well be 4:2:0.

Put differently the HD/SDI link is like a picture frame. You can put a very nice picture in it, or a rather bad one. The frame works the same and doesn't actually alter the picture.

You can get the same result by copying the SxS media to a computer then converting it there to Uncompressed HD 4:2:2.

Just remember- you aren't interpolating or any such. No information is created in this process.

If you need the full 4:2:2 image quality, then the SxS card is useless. You have to record that while the action is happening. You can do that via the HD/SDI link to some sort of outboard recorder.

Does that answer the question properly?

To convert a signal from 4:2:0 to 4:2:2 you have to fill in the missing chroma samples. You can't fill in missing chroma samples with zeros... it would change the chroma signal level from what it should be at this pixel. To fill in the missing chroma you have to calculate the chroma value from the chroma samples in the 4:2:0 signal.

Interpolation (averaging) doesn't mean that you are improving the signal quality... it just means you are filling in the blanks with the best value you can calculate. Essentially you scale up the chroma resolution. Of course this is "empty resolution", not a real increase in resolution.

TV

The HPX500 also uses a lower native resolution (960x540) that gets doubled to 1920x1080

You know, I have yet to actually read anywhere authoritative the actual sensor specifications for the HPX500. Based on he marketing I recieved I thought the HPX 500 was full raster- that's what I get for thinking instead of checking.

Based on the fact that the HPX3000 touts a "full-raster 1920 x 1080" sensor as an "industry first" and your comment I went deeper.

The HPX2000 manual gives its sensor resolution as 1280x720 "valid pixels."

The HPX500 manual doesn't list sensor specifications.

I guess I shouldn't be surprised given that the Varicam is also a 1280x720 sensor. Market differentiation at its very best. (So long as you understand that by "best" I mean "worst.")

It would seem Panasonic has been a bit disingenuous with their marketing materials on the HPX series, which all tout 1080p capability.

Hell, for that matter I can upconvert an XL-1 from its analog outputs- does that make the XL-1 an HD camera? Yeah... didn't think so. If you don't have at least a 1280x720 sensor, you aren't really shooting HD.

Oh, and as to Discovery- I am willing to bet that they won't be accepting HPX500 footage either.

To convert a signal from 4:2:0 to 4:2:2 you have to fill in the missing chroma samples. You can't fill in missing chroma samples with zeros...
::snip::
Essentially you scale up the chroma resolution. Of course this is "empty resolution", not a real increase in resolution.

TV

Tom's being more technical than I am, but saying the same thing. If you have to make a decision though use his explanation.

http://www.dvinfo.net/conf/attachment.php?attachmentid=810&d=1159424266

Here is the image I made that I posted a few times on this forum that show the difference between interlaced 4:2:0 and progressive 4:2:0. This image is being used in a magazine somewhere. I don't have the name at all but I was asked permission to use it. 720p on the Ex1 of course uses the progressive method of chroma sampling. The 1080p modes may use it as well but I do not know for sure yet.

To say progressive 4:2:0 is garbage for keying is just plain silly in my opinion. I am a visual effects artist and I love 4:4:4 RGB but I can get some very good results with progressive 4:2:0. Of course I write my own tools to upsample the chroma channels which helps a lot. By up sampling progressive 4:2:0 I can get a very clean but slightly soft 4:4:4 or 4:2:0. It isn't as detailed but it isn't jagged either. in this case about the worse you will get is maybe a slight darker edge around your subjects which can be dealt with.

One of my earlier points was that if quality is such a huge factor for a heavy visual effects shoot I wouldn't use DVCPROHD at all. In fact I would only use uncompressed in such a situation so the recording format doesn't matter at all. For me the cons of DVCPROHD are too much for that critical of footage. If bandwidth and storage were a concern I would at least capture to Cineform and feel confident that I could maintain the level of quality that I need for my sometimes 20 + layers in a composite. In fact if the quality demands of a shoot are really that high I wouldn't even use HDCAM as I think HDCAM is pretty much the DV of HD. I would take a really good 4:2:0 over 3:1:1 anyday.

Yes, thanks Alexander, it does.

Glad I could help


I guess now I have to do some research on keying and compositing what are essentially 4:2:0 signals.

If you are looking for a workflow I have to ask- are you monitoring footage on set?

If so, then do what I plan on: use an AJA ioHD on set to convert the HD/SDI signal from the camera head, which is real 4:2:2 before the XDCAM compression, into ProRes.

If you have to composite the 4:2:0 signal it is doable... in fact depending on your precise needs it can be straightforward. You can definitely do a good key in 4:2:0

I carry on about 4:2:2 and other trivialities- but you have to remember my work is often premiered in theaters. Some detail that you can't see on a 50" plasma looks huge on a 50' screen.

To convert a signal from 4:2:0 to 4:2:2 you have to fill in the missing chroma samples. You can't fill in missing chroma samples with zeros... it would change the chroma signal level from what it should be at this pixel. To fill in the missing chroma you have to calculate the chroma value from the chroma samples in the 4:2:0 signal.

Interpolation (averaging) doesn't mean that you are improving the signal quality... it just means you are filling in the blanks with the best value you can calculate. Essentially you scale up the chroma resolution. Of course this is "empty resolution", not a real increase in resolution.

TV

It can mean better quality depending on how you want to look at it. If you want crisp details then no it will never be the same. Having smoothed out chroma channels will however give you a much better even key then non upsampled chroma. It should be pointed out that if we want to start nit picking chroma so much that 4:2:2 isn't exactly perfect either. It also has to go through a chroma up sampling in the horizontal direction. Progressive 4:2:0 just means it does the same amount of up sampling in the vertical direction as well.

Progressive 4:2:0 just means it does the same amount of up sampling in the vertical direction as well.

Just???

I agree that 4:2:2 isn't perfect, but its a huge difference from 4:2:0 progressive.

Interesting thread here, but after reading it all I have to say that if utmost quality is needed, then maybe an $8,000 camera is not the correct choice?

If budget is low, then the image quality will have to fall in line.

Okay guys, we've drifted way off the original topic here and have spiraled into a chroma sample measurbating contest.

The notion first put forth in this thread was that given the lower resolution requirements of 720P, the 35 mbit XDCAM HD codec might apply less compression than it would to a 1080 raster. Certainly seems logical and when the camera ships, everyone can do their own testing to see what works for them and what doesn't.

Some interesting information has been put forth and is worth reading.

As a final note, remember that DVINO will not tolerate bashing of anyone, any platform, or any manufacturer. These are all tools and we're lucky to have such variety. Make your choice and exploit its strengths, while minimizing the weaknesses.

Thread is closed for now.

regards,

-gb-