PDA

View Full Version : the whole 4:2:2 thing - to David Heath



Bounce
11-27-2008, 05:57 AM
Hi all. I've been on this forum before but lost my user name and email a long time ago. Big time reader tho.

Okay.




"
David Heath
August 31st, 2008, 09:12 AM

So your point is that the 4:2:2 output might not have much of an impact on the final product compared to the 4:2:0 using the Panasonic cameras?

Do you think keying work would be improved?
"Yes" to the first question above, and "depends" to the second.

Think of "4:2:2" and "4:2:0" recording modes as containers, capable of "holding" resolution (!) Let's say they're each capable of holding 1 litre of luminance resolution :), when 4:2:2 will then hold 1 litre of vertical chrominance information and 0.5 litre of horizontal chrominance information. For 4:2:0 both the chrominance figures are 0.5 litres.

The approach you're suggesting (record HD-SDI via an XDR) effectively replaces a 0.5l container with a 1l container. What I'm saying is that you'll only get an improvement IF the camera head is capable of producing more than 0.5l of vertical chrominance information in the first place. For a camera like the EX the answer is obviously "yes" - it's chips have 1080 pixels vertically each of r, g and b. But for a camera with 960x540 chips, the answers no. Pixel shifting techniques may get about 0.6-0.7l of vertical luminance information (up from the 0.5l you'd expect without pixel shift), but they can't do anything for the chrominance information. You're stuck with 0.5l - double the size of the container, and you just end up with it half empty!

As regards the question of whether keying work will be improved, my guess would be that an XDR type device may have lower overall compression, which could only be a good thing, but the effect of the 4:2:2 recording would be small in this case.

As far as the HPX500 goes, on the up side it's 2/3" chips, and they have proper optical low pass filters to help counter aliasing. (All 1/3" cameras don't, AFAIK). On the other hand, they are 0.5MP chips, so the chrominance resolution won't be as good as a 2MP camera, regardless of the colour space. Colour space numbers are ratios, not absolute numbers, and you have to pay regard to the luminance numbers to make any sense. Would you rather have 50% or 25%? Now what if I asked "50% of 2" or "25% of 8"? In that case 25% is obviously better than 50%, in others 4:2:0 can be better than 4:2:2 by similar logic - you have to make sure like is being compared with like.

I'm afraid that at the end of the day you get what you pay for."





Thus if the 170 has the same chip then converting the 150 MPEG4 file to the DVCAM HD 4:2:2 file (or the cineform 444 for that matter) will it result in the same amount of colour information? Save for the compression degradation?

Thanks

studio1972
11-27-2008, 04:43 PM
Nice post. For what it's worth I was thinking the exact same thing, but you expressed it very clearly.

trez
11-28-2008, 08:13 AM
Thus if the 170 has the same chip then converting the 150 MPEG4 file to the DVCAM HD 4:2:2 file (or the cineform 444 for that matter) will it result in the same amount of colour information? Save for the compression degradation

What a coincidence... I wrote a post on this in another thread yesterday... An interesting conclusion is that the DVCPRO HD codec's 4:2:2 chroma is an overkill for HPX170 - lot of the chroma information being carried by the codec is redundant, because of the limited chip resolution. This not a big problem considering enormous 100Mbit/s DVCPRO HD throughput. But... Things get worse when shooting 720p - DVCPRO HD bitrate becomes 40Mbit/s then. No surprise HMC150 with its increased raster size, 4:2:0 chroma (better bitrate usage) and advanced MPEG4 compression outperforms HPX170 in 720p.

PerroneFord
11-28-2008, 08:33 AM
Well this should get quite interesting...

reem12
11-28-2008, 08:43 AM
Not realy buying this 150 out performing the 170 thing. There is clearly a difference in overall color rendition from both cams. IF someone can post a 720 heavy color corrected still of both cams i'd love to see.

I would not even say the 150 out performs the a1 when factoring the a1s native resolution and glass quality, and especialy after you have dialed in one of stevens presets.

trez
11-28-2008, 09:14 AM
Not realy buying this 150 out performing the 170 thing. There is clearly a difference in overall color rendition from both cams. IF someone can post a 720 heavy color corrected still of both cams i'd love to see.

Didn't have the chance to compare them side-by-side, but I can't think of any reason why there's a difference in color rendition. I don't mean there isn't any. I'm just trying to find out why - everything, but codec, is the same in both cameras - glass, sensors, processing chip. Both AVCHD and DVCPRO HD are 8-bit color... The only difference seems to be the chroma downsampling factor, but, as I said, 4:2:2 is an overkill for these cameras (my opinion). Maybe there's something in the encoding algorithm itself that ruins the color information - that would be very strange, though.
Or... Panasonic does something (on purpose) to make the HMC150 footage look worse ;)

I realize cc can reveal differences in more apparent way. But I don't really think comparing heavily cc-ed footage is the right procedure unless the original codec (prior to cc) is the same. Probably the most fair way would be transcoding the footage from 170 and 150 to uncompressed, then applying cc.Wish I had both cams at home ;)

Bounce
11-30-2008, 04:25 PM
no answer?

i'l recap: if the 150/170 produces colour (chrominance) information that can not fill the 4:2:2 codec - therefore only 4:2:0 is necessary.

THUS the only reason for the 4:2:2 DVCPRO HD codec is its lighter compression ratio (less artifacting and possibly a touch more detail).

As Trez alluded to, the best test would be to transcode the footage from both cameras to the Cineform 4:4:4 codec and see what colour control you have. A further test would be to transcode only the 150's footage into DVCPRO HD and see how much you lose compared to the 170's native DVCPRO HD.

Theoretical excersise? Maybe. But it will prove that maybe the DVCPRO HD codec is only good for it's compression ratio and not its colour space (in this case only of course). Plus, I might just buy the 150 instead of the 170 and lose a bit on the compression (no biggie).

I'm no expert, i'm just talking. If someone really knows the answer to this it would be great to know.

Cheers

Spartacus
11-30-2008, 05:05 PM
I´m hooked on...

David Jimerson
11-30-2008, 05:17 PM
I don't see how HMC150 footage could be the same as HPX170 footage if you converted them both to the same codec -- they have the same optics, and even if there weren't "enough" color to fill 4:2:2, the HMC150 still throws out twice as much of the color which IS there as the 170 does. It IS a ratio, and if you're comparing apples-to-apples signals between footage coded to 4:2:0 and to 4:2:2, 4:2:2 has twice the color info. That Heath thing looks to me like someone rationalizing why, in theory, the EX1 should still be better at color than the 200 or 170 even though it's 4:2:0 and not 4:2:2.

If you want to see the color difference between the 150 and the 170, just try pulling keys from both.

Bounce
11-30-2008, 06:06 PM
so the 150 removes half the colour that the 170 does?

do you mean the chips produce the same amount of colour information, but the 150 keeps only 50 % of that?

OR are you saying that the DVCPRO HD codec simply doubles whatever is there (using its own algorithm)?

David Jimerson
11-30-2008, 06:23 PM
http://www.dvxuser.com/articles/colorspace/

Bounce
11-30-2008, 06:33 PM
read it. doesn't answer question.

plus I don't think David Heath is pulling our legs so we can buy an EX cam.

if David is correct then the link http://www.dvxuser.com/articles/colorspace/ (http://www.dvxuser.com/V6/../articles/colorspace/) simply backs up what he says.

David Jimerson
11-30-2008, 07:15 PM
When you take the same signal and encode it as 1) AVCHD and 2) DVCPRO HD, the DVCPRO HD retains twice as much color resolution (and accuracy, as the article points out). To say nothing of the level of compression.

Bounce
11-30-2008, 07:28 PM
So it RETAINS (does not create...) twice the amount of colour information.

Cool. Thanks for answering that, as the article you mention does not mention the 150 or the 170 in its discussion of colour space and pixels (though we all know about their codecs).

PerroneFord
11-30-2008, 07:58 PM
We've been down this path before. If we stipulate to the HVX/HPX recording in DVCProHD, then we have a 1280x1080 signal laid down.

4:2:2 extracted from that 1280x1080 signal is VERY close to 4:2:0 extracted from a full raster 1920x1080. The 4:2:2 signal maintains an edge, but only slightly.

But the whole issue is pointless. If you REALLY need the color info, just come off the SDI port and call it a day.

shrigg
11-30-2008, 08:26 PM
Or the HDMI port.

Bounce
11-30-2008, 11:18 PM
SDI or HDMI? I'm confused.

So, David Heath is clearly wrong in saying that.....

"The approach you're suggesting (record HD-SDI via an XDR) effectively replaces a 0.5l container with a 1l container. What I'm saying is that you'll only get an improvement IF the camera head is capable of producing more than 0.5l of vertical chrominance information in the first place. For a camera like the EX the answer is obviously "yes" - it's chips have 1080 pixels vertically each of r, g and b. But for a camera with 960x540 chips, the answers no. Pixel shifting techniques may get about 0.6-0.7l of vertical luminance information (up from the 0.5l you'd expect without pixel shift), but they can't do anything for the chrominance information. You're stuck with 0.5l - double the size of the container, and you just end up with it half empty!"


I still want a 150 - pity it doesn't have auto audio gain......

PerroneFord
12-01-2008, 12:17 AM
I'm not saying he's wrong at all...

Spartacus
12-01-2008, 02:24 AM
BTW where is the original David Heath post from?

Bounce
12-01-2008, 05:33 AM
here in DVX land

BobDiaz
12-01-2008, 09:43 AM
In reading several of the messages from the last few days, I get the feeling that some people are looking at the numbers and not considering how the image looks.

First, if you are doing any blue or green keying, yes higher color resolution is better. That would give a better edge on your key. However, 4:2:0 keying is still possible.


Assuming that you are not doing any chroma keying, if you look at the videos and stills posted from the HMC-150, the results look really good.

Our eyes are less sensitive to color detail than to black & white detail. This is why color resolution is almost always lower than the luminance (black & white) resolution. So, yes on paper, the 170 does produce higher color resolution than the 150, BUT in a wedding video, business video, or some other video, it's highly unlikely that you or your viewers are going to see the difference.

Both Broadcast TV and Blu-ray transmit the video as 4:2:0. So, the end result won't be able to show the real difference.

The reason for selecting a 170 over the 150 isn't color resolution, but work flow or extra features.


Bob Diaz

trez
12-01-2008, 11:47 AM
In reading several of the messages from the last few days, I get the feeling that some people are looking at the numbers and not considering how the image looks.
First, if you are doing any blue or green keying, yes higher color resolution is better. That would give a better edge on your key. However, 4:2:0 keying is still possible.

Bob, I'm sorry to say that, but I have the feeling you haven't read those messages carefuly and are missing the point.
Of course 4:2:2 is better than 4:2:0. Nobody says it is not.
What plenty of people say is that HPX170 does not deliver true 4:2:2. Its codec's downsampling factor is 4:2:2 but it carries an upscaled chroma (thus a lot of redundant information).

How will you explain the fact, that HPX170 sensors deliver 960x540 chroma resolution while in the final DVCPRO HD footage (1080) the chroma resolution is 640x1080. Where do these additional 540 vertical chroma samples come from? 960x540 chroma samples is enough to carry all the information the CCDs capture - everything above this is simply an overkill.

The conlusion is that DVCPRO HD codec is not optimal for the HPX170 camera construction. It has been used in this camera because it is a standard - good standard! But this camera is simply not capable of exploiting its full potential (though it is capable of delivering great looking image, of course). This is what this discussion is about.

Bounce
12-01-2008, 02:21 PM
Bob,

For me, I would rather a camera that does the full 4:2:2 cos I colour grade and composite my footage.

...though I don't do green screen, I do however isolate colours and change them to something else.

Plus a 4:2:2 camera and codec is more colour accurate.




On another point, that XDR capture device would theoretically give you a better image than the 170's DVCPRO codec. Plus 24 bit sound. Plus varying frame rates (I think). Thus a 150 with the XDR may turn out to yield a better image than a 170 on its own - we shall see.

David Jimerson
12-01-2008, 02:24 PM
Both Broadcast TV and Blu-ray transmit the video as 4:2:0. So, the end result won't be able to show the real difference.

That is true, but there's a big step in between acquisition and the end result where it's pretty important. :)

Jan_Crittenden
12-01-2008, 02:53 PM
Bob, I'm sorry to say that, but I have the feeling you haven't read those messages carefuly and are missing the point.
Of course 4:2:2 is better than 4:2:0. Nobody says it is not.
What plenty of people say is that HPX170 does not deliver true 4:2:2. Its codec's downsampling factor is 4:2:2 but it carries an upscaled chroma (thus a lot of redundant information).

The 4:2:2 is from of the DVCPRO Codec, that codec is the same as the codec that is on the Varicam or the HPX2000. It is a 4:2:2 and there is nothing about the chip set that can change this. Those plenty of people that say otherwise, well they are just confused by marketing and not engineering.


How will you explain the fact, that HPX170 sensors deliver 960x540 chroma resolution while in the final DVCPRO HD footage (1080) the chroma resolution is 640x1080. Where do these additional 540 vertical chroma samples come from? 960x540 chroma samples is enough to carry all the information the CCDs capture - everything above this is simply an overkill.

The Sensors on the 170, 150, and the 200/200A are spatially offset to deliver to the DSP an effective 1.1 million pixels by virtue of the spatial offset. That DSP captures the signal at a 1080P capture, all the processing is in the 1080P domain and once the camera processing is over, the Codec comes into play. DVCPRO HD is a 4:2:2 codec which means that there are 2 samples of red and 2 samples of blue for every luminance sample. It also means that there is color on every line.



The conlusion is that DVCPRO HD codec is not optimal for the HPX170 camera construction. It has been used in this camera because it is a standard - good standard! But this camera is simply not capable of exploiting its full potential (though it is capable of delivering great looking image, of course). This is what this discussion is about.

Are there other cameras that have more resolution, yes, but to say that the 170 doesn't exploit the virtures of the DVCPRO HD codec is simply incorrect. Sorry. An I-Frame codec is superior to a Long GOP any day of the week. That is the simple truth.

All the best,

Jan

PerroneFord
12-01-2008, 03:15 PM
... Those plenty of people that say otherwise, well they are just confused by marketing and not engineering.

An I-Frame codec is superior to a Long GOP any day of the week. That is the simple truth.


What's that you were saying about marketing and not engineering?

I-Frame is superior to Long GOP eh? So my Mini-DV is superior to XDCam? And DVCProHD is superior to HDCamSR?

Right.

Barry_Green
12-01-2008, 03:33 PM
HDCAM SR is not long-GOP. HDCAM SR is based on MPEG-4, but they use only an intraframe profile. Just like how Sony's own MPEG-IMX is an MPEG-2 implementation but they made it intraframe-only.

I'm certain that Jan was referring to like-for-like, so DV would not be compared to XDCAM (although, of course, you could, since XDCAM uses the exact same DV format/compression as your mini-DV does!) Certainly you meant to say XDCAM-HD or XDCAM-EX, right? But if you want to play "gotcha", you could be "gotcha'd" just like that.

HDCAM SR competes with D5-HD, not with DVCPRO-HD (and yes, they're all I-frame-only). DVCPRO-HD competes against HDCAM, and HDCAM is an I-frame-only codec. Of course, so is every codec Sony ever made for professionals, up until this recent MPEG-2 fascination. DigiBeta, DVCAM, DV, HDCAM, HDCAM-SR, MPEG-IMX, they're all intraframe-only.

BobDiaz
12-01-2008, 03:54 PM
Correct me if I'm wrong about this, but I'm not aware of any HDTV camera for under $10,000 (USA price) that captures a full 4:2:2 color. Even the Sony EX1 & EX-3 are 4:2:0 color. Both the Canon A1 and H1 are 4:2:0 color.


In the case of the HMC-150, the color is recorded 4:2:0, looking at the color information for each row, first point, it comes together like this (starting at row 100):

Row 100: Color ((Row100+Row101)/2)
Row 101: Color ((Row100+Row101)/2)

Row 102: Color ((Row102+Row103)/2)
Row 103: Color ((Row102+Row103)/2)

...

In the case of the HVX200[a] and the HPX-170, the color is recorded as 4:2:2. Looking in the vertical direction and starting at Row 100, the color would be as follows:

Row 100: Green Pixel Row 100, Red & Blue from (Row 99+Row 101)/2
Row 101: Green from (Row 100+Row102)/2, Red & Blue Row 101
Row 102: Green Pixel Row 102, Red & Blue from (Row 101+Row 103)/2
Row 103: Green from (Row 102+Row104)/2, Red & Blue Row 103
...


The 4:2:2 of the 170 and the 200[a] does provide some color change information for each row, but is NOT as good as a more expensive 2/3" camera that has the full 1080 vertical pixels for RGB.



Bob Diaz

Jan_Crittenden
12-01-2008, 03:58 PM
What's that you were saying about marketing and not engineering?

I-Frame is superior to Long GOP eh? So my Mini-DV is superior to XDCam? And DVCProHD is superior to HDCamSR?

Right.

Come on Perrone, You know what I meant. And if we were comparing cameras at the same level like DVCPRO 50 and the XDCAM, yeah the DVCPRo50 would be better. Since there isn't an SD $5000 camera or even a $3000 that does XDCAM, it would be hard to say if it would be better. But here we are talking about 1/3" cameras and the diff between long GOP recording codecs and DVCPRO HD, 4:2:0 vs. 4:2:2. and I Frame and 4:2:2 win out.

This isn't marketing, it is engineering.

Best,

Jan

Jan_Crittenden
12-01-2008, 04:00 PM
Correct me if I'm wrong about this, but I'm not aware of any HDTV camera for under $10,000 (USA price) that captures a full 4:2:2 color. Even the Sony EX1 & EX-3 are 4:2:0 color. Both the Canon A1 and H1 are 4:2:0 color.

Bob, you are wrong.

Best,

Jan

trez
12-01-2008, 04:22 PM
The 4:2:2 is from of the DVCPRO Codec, that codec is the same as the codec that is on the Varicam or the HPX2000. It is a 4:2:2 and there is nothing about the chip set that can change this. Those plenty of people that say otherwise, well they are just confused by marketing and not engineering.

Jan - everyone here agrees that DVCPRO HD is 4:2:2.


The Sensors on the 170, 150, and the 200/200A are spatially offset to deliver to the DSP an effective 1.1 million pixels by virtue of the spatial offset.

Of course. We all know how pixel-shifting works. It is capable of resolving more LUMA samples than the sensor nominal resolution (1.1 million being an optimistic scenario - it all depends on the image being captured).
However - this has nothing to do with CHROMA!
F.ex. - there's only one sensor capturing red. Its resolution is 960x540. You won' get more samples by combining it with either green or blue sensor output, because they are not sensitive to red at all.
With luma it's very different, because each luma sample is weighted average of three channels - this is why pixel shifting magic works. Each one of the sensors can contribute to luma. Offsetting the green sensor results in the ability to resolve more luma samples.


That DSP captures the signal at a 1080P capture, all the processing is in the 1080P domain and once the camera processing is over, the Codec comes into play.

Of course. This is obvious.


DVCPRO HD is a 4:2:2 codec which means that there are 2 samples of red and 2 samples of blue for every luminance sample.

It's the opposite, actually (2 luminance samples for every chroma sample). Just a typo.


Are there other cameras that have more resolution, yes, but to say that the 170 doesn't exploit the virtures of the DVCPRO HD codec is simply incorrect. Sorry.

Nobody said 170 does not exploit the virtues of the DVCPRO HD codec. I said it did not exploit the full potential of ONE of the codec properties - chroma resolution. Simply because its sensors don't resolve as many chroma samples as the codec can carry. So, some part of the chroma carried by the codec is the result of in-camera math.

Please don't over-interpret my posts. I am Panasonic user and I have good reasons to be one. In my opinion 170 is a better camera than 150! What I attempt to do is to reveal the reasons why it is superior. And 4:2:2 chroma space is not one of them, IMO. I dare to claim you can get the same chroma quality by smoothing the 150 4:2:0 chroma (compression artifacts aside).
Of course I may be wrong. Still waiting for arguments ;)


An I-Frame codec is superior to a Long GOP any day of the week. That is the simple truth.

This is simple truth. To be precise - it all depends on the bitrate and how long the group of picture is... But in the real world implementations the I-Frame codecs are superior to Long GOP ones. So I agree with this one.

Don't get me wrong. I'm not trying to say HPX170 is bad camera. I'm not trying to say that using DVCPRO HD in this camera was wrong decision - it was great!
I'm only trying to say two things:
1. Some of the 170's DVCPRO HD bitrate carries redundant information. Not much, though. Just the tiny fraction of the huge 100Mbit/s bandwidth. Don't get me wrong - the camera exploits the DVCPRO codec as much as it can. It just can exploit it fully due to limited sensor resolution.
2. 170s 4:2:2 chroma is not an advantage over 150's 4:2:0 chroma. When hesitating whether to buy 150 or 170 take other reasons into accounts (and there are MANY)

PerroneFord
12-01-2008, 04:40 PM
HDCAM SR is not long-GOP. HDCAM SR is based on MPEG-4, but they use only an intraframe profile. Just like how Sony's own MPEG-IMX is an MPEG-2 implementation but they made it intraframe-only.


About 10 minutes after I posted that, I realized that Mpeg4 does have the intraframe mode like AVCIntra. So I have no problem being corrected there.



I'm certain that Jan was referring to like-for-like, so DV would not be compared to XDCAM (although, of course, you could, since XDCAM uses the exact same DV format/compression as your mini-DV does!) Certainly you meant to say XDCAM-HD or XDCAM-EX, right? But if you want to play "gotcha", you could be "gotcha'd" just like that.


Yea, I could be gotcha'd.



HDCAM SR competes with D5-HD, not with DVCPRO-HD (and yes, they're all I-frame-only). DVCPRO-HD competes against HDCAM, and HDCAM is an I-frame-only codec. Of course, so is every codec Sony ever made for professionals, up until this recent MPEG-2 fascination. DigiBeta, DVCAM, DV, HDCAM, HDCAM-SR, MPEG-IMX, they're all intraframe-only.

The thing is, that these threads get so frustrating because of the spin. Probably best if people like myself stay out of them.

I have 2 Panasonic cameras, 2 Canons, and one Sony. They all make me and my clients happy. I'll just leave it at that.

Jan_Crittenden
12-01-2008, 04:56 PM
Sure - everyone agrees that DVCPRO HD is 4:2:2.
Of course. We all know how pixel-shifting works. It is capable of resolving more LUMA samples than the sensor nominal resolution (1.1 million being an optimistic scenario - it all depends on the image being captured).
However - this has nothing to do with CHROMA!

Actually I am not convinced that everone knows how spatial offset works because thes discussions come up all the time. I have a known 1.1 million pixel imager camera in my lab. I compare it to the HVX200/170. Resolution is virtually the same. Yet everyone looks at the numbers on the chipset and just disses the camera's ability to make competitive pictures or challenge the DVCPRO codec. as you put it. Heck for all you know it may be the lens that is getting in the way of using the entire DVCPRO HD codec. The difference is that the signal coming off of the chips is analog signal, there is not a pixel to pixel relationship to the recording.



F.ex. - there's only one sensor capturing red. Its resolution is 960x540. You won' get more samples by combining it with either green or blue sensor output, because they are not sensitive to red at all.

But there are nuances in other colors say purple that are made up from the red and blue sensor. If there is any tonal gradation, you just added green. There is more resolution to be had. No if the image was only red okay, but how many times is the image only red? Few I would hope, and from my experience that is true, pretty rare for an all red movie. I have looked at spatial offset and it is probably one of the leading technologies out there, it does make something out of what seems to be nothing. We use it in every single camera we make, including the 3000, which is a 2 million pixel imager. Why, because it adds something.



With luma it's very different, because each luma sample is weighted average of three channels - this is why pixel shifting magic works. Each one of the sensors can contribute to luma. Offsetting the green sensor results in the ability to resolve more luma samples.

It is and offset of the red and the blue not the offset of the green. It would make it vastly more difficult as 70% of the image in HD comes from the Green channel.




It's the opposite, actually (2 luminance samples for every chroma sample). Just a typo.

OOps, or 2 chroma red and 2 chroma blue for every 4 green with every line having color.


Please don't over-interpret my posts.

I am only reading your posts as an novice might. And they might be lead down a very different path than the one you are stating now.


I am Panasonic user and I have good reasons to be one. In my opinion 170 is a better camera than 150!

I would agree, but that is because of the DVCPRO HD codec, Intra-frame and 4:2:2.


What I attempt to do is to reveal the reasons why it is superior. And 4:2:2 chroma space is not one of them, IMO. I dare to claim you can get the same chroma quality by smoothing the 150 4:2:0 chroma (compression artifacts aside).
Of course I may be wrong. Still waiting for arguments ;)

Seen it, you can't make 4:2:0 into 4:2:2. Dosen't matter how good the camera is, there is a dead giveaway in the fact that every other line has color.


I'd say it all depends on the bitrate and how long the group of picture is... Of course in the real world implementations most I-Frame codecs are superior to Long GOP ones. So I agree with this one.

I don't think it depends on the bitrate. Seems to be evident in very green screen I have seen for any of the 4:2:0 codecs vs, the 4:2:2 ones.



1. Some of the DVCPRO HD bitrate is used to carry redundant information. Not much, though. Just the tiny fraction of the huge 100Mbit/s bandwidth.

I wouldn't agree with this.


2. 170s 4:2:2 chroma is not an advantage over 150's 4:2:0 chroma. When hesitating whether to buy 150 or 170 take other reasons into accounts (and there are MANY)[/QUOTE]

If you are doing green screen work, it most definitely is.

All the best,

Jan

kurtmo
12-01-2008, 05:42 PM
I got decent results keying with a 150. Results (http://vimeo.com/1900735). Then again, I'm comparing to the results I got with a DVX100b.

It'd be nice to have someone take some green screen footage with a 150 and a 170 side-by-side and see how they compare.

Jan_Crittenden
12-01-2008, 05:55 PM
I got decent results keying with a 150. Results (http://vimeo.com/1900735). Then again, I'm comparing to the results I got with a DVX100b.

It'd be nice to have someone take some green screen footage with a 150 and a 170 side-by-side and see how they compare.


Exactly. AVCHD is to HD what DV was to SD. If your budget is <$4000 the the results you get with the HMC150 will be better than any other long GOP 4:2:0 codec as the AVCHD codec is a much better codec. When we start talking about the HPX170 and his brother, well we are talking about a camera that costs $2000 more and then there are the P2 cards.

Your result is exactly what I have seen and the differences are in the edges of objects and singular hairs blowing in the wind. Everyone has a budget and if the HMC150 is your budget, you have a fabulous little camera! More money, more better.

All the best,

Jan

Noel Evans
12-01-2008, 06:16 PM
H
As far as the HPX500 goes, on the up side it's 2/3" chips, and they have proper optical low pass filters to help counter aliasing. (All 1/3" cameras don't, AFAIK). On the other hand, they are 0.5MP chips,

Just popped in to make a small correction. The 500 chips are not 0.5mp. Thats all.

Bit too much number crunching going on here. If you are doing CRITICAL green screen work then I wouldnt use a 150 or a 170. But for most application, both would be fine. The canon A1 does a good job, so stands to reason the 150 would as well .

Bounce
12-02-2008, 12:35 AM
here in DVX land

Hello.

A correction. The original post was sourced from a google search which gave me an archived forum thread. Because it was about the Panasonic I erroneously claimed it was from dvxuser.com

The original quote was actually from dvinfo.net. See here:

http://www.dvinfo.net/conf/panasonic-avccam-camcorders/127098-press-release-panasonic-unveils-hmc150-pricing-ship-date-5.html

Also, before anyone starts a debate about the Z1 image quality compared to the 150; it is by now (the original thread is before the release of the 150) clear that the 150 has a significantly better image.


Thanks

trez
12-02-2008, 12:49 AM
Actually I am not convinced that everone knows how spatial offset works because thes discussions come up all the time.

I read your post and realized I hadn't known it either ;)
You are right - the image is not just plain green or red, but also all the nuances. So, the output of the different subpixel-shifted sensors, sampled at the frequency higher than nominal sensor nominal resolution can resolve that extra chroma information. This is because the photoactive regions overlap (this is true for green/red and green/blue). This actually increases the overall chroma resolution beyond that of single sensor. It's exactly the same process that allows for greater luma resolution.
Thanks for explaining this, Jan! It is simple once you know it, but nobody has been able to clarify this before.


I have looked at spatial offset and it is probably one of the leading technologies out there, it does make something out of what seems to be nothing. We use it in every single camera we make, including the 3000, which is a 2 million pixel imager. Why, because it adds something.

I've always understood how this improves the luma resolution. But it wasn't clear how it affected the chroma. Maybe it's because all the articles about pixel shifting focus on the luminance resolution.
So, because it is the green sensor that is offset biaxially (relative to red/blue), the camera is capable of resolving finer details of, f.ex. the edge between any greenish (greenscreen) area and anything of different color. This will definitely improve greenscreen compositing. Just another reason why greenscreen is more suitable for video than bluescreen! (the pixel-shifting won't work for pure red object on pure blue area).

Now, the important question arises: Does HMC150 HDMI output the real 'pixel-shifting driven' 4:2:2 chroma or is it just the plain 150 4:2:0 chroma upsampled to 4:2:2?

Bounce
12-02-2008, 01:00 AM
HDCAM SR is not long-GOP. HDCAM SR is based on MPEG-4, but they use only an intraframe profile. Just like how Sony's own MPEG-IMX is an MPEG-2 implementation but they made it intraframe-only.

I'm certain that Jan was referring to like-for-like, so DV would not be compared to XDCAM (although, of course, you could, since XDCAM uses the exact same DV format/compression as your mini-DV does!) Certainly you meant to say XDCAM-HD or XDCAM-EX, right? But if you want to play "gotcha", you could be "gotcha'd" just like that.

HDCAM SR competes with D5-HD, not with DVCPRO-HD (and yes, they're all I-frame-only). DVCPRO-HD competes against HDCAM, and HDCAM is an I-frame-only codec. Of course, so is every codec Sony ever made for professionals, up until this recent MPEG-2 fascination. DigiBeta, DVCAM, DV, HDCAM, HDCAM-SR, MPEG-IMX, they're all intraframe-only.


So, HDV is MPEG-2 and thus long-GOP at 25 megabits.

DVCPRO HD (HVX) is intraframe and 100 megabits.

Would then an MPEG-2 long-GOP at 100 megabits have more info than the DVCPRO HD intraframe?

I ask the more knowledgable - I'm here to learn - please be gentle.

Cheers

ilauzirika
12-02-2008, 01:35 AM
In the, what matters is how the images look to the eyes of viewers.... about greenscreen....green edges are legit, have you recently seen star wars (the good orginal ones) they are full of green and blue edges.... if those films have edges, edges are the way to go..... I know that they are really old but it doesn't mind...... if you want some greenscreen footage look in the footage subforum.

Bounce
12-02-2008, 03:39 AM
Hi. I actually contact David Heath regarding this thread. I don't know if he has a user account but he has written to me and has given me permission to quote his emails.

Here they are:


email 1

"As far as the DVXuser posts in that thread go, then a couple of points. The whole 4:x:x notation should really be only used in relation to a RECORDING system, and it starts to get very messy when it gets used more widely. The important essence is that it denotes the ratio of recorded colour samples to luminance samples, just as the article that David Jimerson posted to in that thread says. What’s important in this context is what happens in the vertical direction, and that thread is assuming 1080 video, so we’ll consider a vertical line of 1080 pixels.

In this case, 4:2:2 implies that 1080 luminance and 1080 chrominance samples will be recorded vertically, 4:2:0 implies that 1080 luminance and 540 chrominance samples will be recorded. The numbers say NOTHING about what the front end of the camera is generating. The assumption in the DVXuser article is that the head output has equal luminance/chrominance resolution, and for a non-pixel shift 3 CCD camera that’s true. But cameras with Bayer sensors, or those using pixel shift, have a luminance resolution that is inherently greater than the chrominance resolution. I stress there is nothing wrong with that. It matches the human eye, which is inherently more sensitive to changes in brightness than in hue, and is the whole reason why colour subsampling is used in broadcasting.

In the case of the 200/171/151, there are 540 r,g,b pixels each vertically, so at first sight the luminance and chrominance resolutions may be thought to be 540 (lens permitting). In practice, the r and b pixels are offset from the g to enhance the luminance resolution (by a factor of about 1.5X overall, and because the camera uses 2 dimensional shift, it’s about sq root 1.5 on each axis, say 1.25x) but this doesn’t do anything for the chrominance resolution. That 1.5x figure comes straight from the Panasonic Technical White Paper, incidentally, ( ftp://ftp.panasonic.com/pub/Panasonic/Drivers/PBTS/papers/AG-HVX200.CCD-WP.pdf ) at page 3: “Spatial Offset technology will improve the resolution by a factor of 1.5 provided the gain in resolution is not offset by the quality of the lens.”

Hence although DVCProHD is capable of RECORDING 1080 chroma samples vertically (it’s a 4:2:2 recording), the camera front end can only supply it with information corresponding to 540. This was my analogy to a container half empty. (If anyone disputes that, ask them what happens when confronted by a blue lines on black test chart, such that only the blue pixels produce an output. It CAN’T resolve better than 540 vertically.) Now take a camera such as the EX which has 1080 r,g,b pixels vertically, and here the front end IS capable of supplying 1080 chroma samples. Unfortunately, the 4:2:0 subsampling throws every other one away, so again the nett result is 540 RECORDED chroma samples. Hence the analogy of trying to pour a litre into a 0.5 litre pot, half gets spilt. Far from “rationalizing why, in theory, the EX1 should still be better at color than the 200 or 170” as David Jimmerson suspects, I’m actually saying that they both have more or less the SAME vertical chroma resolution RECORDED, 540! (Though the EX is capable of supplying better via HD-SDI.)

Now move on to the 151, and here the 4:2:0 recording of AVCCAM means it has the capacity to record 540 vertical chroma pixels – which matches nicely what the front end is capable of. Hence all three examples end up with effectively 540 vertical res recorded.

It’s a different topic, but it’s worth thinking about recording 720p instead of 1080. The 151 720 codec records full raster 1280x720 in this mode, DVCProHD records 960x720. The resolution of the front end should be around (960x1.25)x(540x1.25) from theory (about 1200x650), which matches an AVCCAM 720 raster very well. Record to 720p DVCProHD and quite a lot of horizontal detail is lost. Hence for a 151, 720 recording should preserve as much luminance detail as 1080, whilst giving the compressor a much easier time.

What isn’t clear from the original quote at the head of the thread is that in the original DVinfo thread it came from I was replying to a question about how effective it would be to use a CV XDR with a 151 (see post 62 of the original thread), and was addressing a very specific circumstance rather than a general 422/420 one. Take that into consideration, and what was quoted makes more sense.

At the end of the day, all this is only one factor in making a camera choice, though it’s good to understand the technicalities. Personally, I think the SDHC card ability of the HMC151 versus P2 of the 171 is a far, far more significant factor than nit-picking over resolution numbers, and the extra difficulty of editing AVCHD can be overcome by transcoding on import. If you’re looking for a “B” camera to an otherwise 2/3” P2 shoot, the 171 and P2 may be the way to go, otherwise the advantages of SDHC seem to make the 151 a better choice than a 171 for the vast majority of people.

Hope that helps answer your query,

David"

Bounce
12-02-2008, 03:42 AM
email 2

"Yes, please do reproduce my reply to you if you wish. I’ve looked again at that thread since sending you the reply, and note that quite a lot extra has been subsequently said, and we seem now to have a Panasonic marketing person contradicting a Panasonic White Paper! (The one referred to in my first e-mail to you.) The no of pixels involved for each colour are 518,400 (960x540), and it’s true to say that red, blue give different positional information to green, hence a total of 1,036,800 potentially different sites. This I believe to be where a 1 megapixel figure comes from, I’m not sure about 1.1MP?

But it’s wrong to then say it equates by pixel shift processing to a 1 megapixel luminance sensor for two main reasons. Firstly, unlike a true 1MP sensor, the r/b and g pixels optically overlap, and secondly that for peak white stimulation the r/b or g sites will only give out half a corresponding luminance unit. It’s generally accepted (including in the Panasonic white paper) that the improvement factor therefore comes down to about 1.5x the non pixel shifted chip dimensions, so the 0.5MP chips here used get a boost to the order of about 0.75 MP equivalent, far from insignificant. Because the effect is spread equally between the H&V axis, that makes it equivalent to about 1200x650 in resolution terms (1200x650=780,000). There is actually a lot more to it than that, such as trade offs of diagonal resolution for hor/vert, aliasing issues, etc but the above is a good first approximation.

As another poster stresses, this is for luminance only. Ask yourself what the result would be of a series of blue/black horizontal lines, 540 blue, 540 black? The DVCProHD codec could certainly record the pattern if it was presented, but vertically, each camera chip pixel would be presented with one black, one blue, so the chip certainly couldn’t resolve it. Go to in between situations, (pink, pale yellow lines, say) and the system may well then resolve more than 540, but nearly all the extra information will then be in the luminance recording channel anyway.

Hope all this makes sense.

David"

Jan_Crittenden
12-02-2008, 07:56 AM
That 1.5x figure comes straight from the Panasonic Technical White Paper, incidentally, ( ftp://ftp.panasonic.com/pub/Panasonic/Drivers/PBTS/papers/AG-HVX200.CCD-WP.pdf ) at page 3: “Spatial Offset technology will improve the resolution by a factor of 1.5 provided the gain in resolution is not offset by the quality of the lens.” Hence although DVCProHD is capable of RECORDING 1080 chroma samples vertically (it’s a 4:2:2 recording), the camera front end can only supply it with information corresponding to 540.

But he is missing the point on the entire picture here. It is breaking down to a level that it becomes confusing because it is taking the numbers out of its context and asking each to operate in isolation. This is what happens when you look at 540 on this looks like 540 on that I mean yeah 540 is 540, but how does that 540 relate to the rest of the codec, its prefilter and final compression. It is a complex thing and for the most part, it seems like a fruitless way of comparing unless the only thing you want to prove is numbers. Pictures on the other hand get you paid.


This was my analogy to a container half empty. (If anyone disputes that, ask them what happens when confronted by a blue lines on black test chart, such that only the blue pixels produce an output. It CAN’T resolve better than 540 vertically.)


The Blue in HD makes up 7% of the HD signal and as with the red anology above, not much is all the right color blue that will not be a combination of colors and thus lend a hand to the luminance resolution which is 70% of the picture. Blue in and of itself makes such a small contribution it is hardly worth noting. Red is only 23%.


Now take a camera such as the EX which has 1080 r,g,b pixels vertically, and here the front end IS capable of supplying 1080 chroma samples. Unfortunately, the 4:2:0 subsampling throws every other one away, so again the nett result is 540 RECORDED chroma samples.

What David seems to be missing is that the 4:2:2 denotation is a ratio. It isn't a matter of how many as compared from one camera to another, it is a ratio of how many as compared to its own luminance sample. Thus it is a 4:2:0 signal with every other line having no color at all. 4:2:0 has color on every other line. That is engineering.


Hence the analogy of trying to pour a litre into a 0.5 litre pot, half gets spilt. Far from “rationalizing why, in theory, the EX1 should still be better at color than the 200 or 170” as David Jimmerson suspects, I’m actually saying that they both have more or less the SAME vertical chroma resolution RECORDED, 540! (Though the EX is capable of supplying better via HD-SDI.)

Well it is not true that the recordings of the EX1 has better color than that of the HVX. It may be that the recoding coming off of the HD-SDI into DVCPRO HD would be better but never on an HDV nor on XDCAM-EX recording.


Now move on to the 151, and here the 4:2:0 recording of AVCCAM means it has the capacity to record 540 vertical chroma pixels – which matches nicely what the front end is capable of. Hence all three examples end up with effectively 540 vertical res recorded.

This is the same sort of thing that I have been saying since I joined this thread. 540 is not the maximum number of resovling points on any of the 540X960 imagers and this is due to the offset. Even though the guy references the document he clearly does not understand it. This feels like a bunch of smoke as this person has clearly had no hands on experience with the camera nor the AVCHD codec. It also seems pretty apparent that he has sat through one too many Sony presentations. ( I say that because of the silly agrument of the chroma numbers comparison above, that is a biggy in their presentation, and they just fail to denote that the designation is a ratio of color samples to its own luminance) The more luminance resolution to cover with the color the more spread out the count becomes. It is a ratio to itself.



At the end of the day, all this is only one factor in making a camera choice, though it’s good to understand the technicalities. Personally, I think the SDHC card ability of the HMC151 versus P2 of the 171 is a far, far more significant factor than nit-picking over resolution numbers, and the extra difficulty of editing AVCHD can be overcome by transcoding on import. If you’re looking for a “B” camera to an otherwise 2/3” P2 shoot, the 171 and P2 may be the way to go, otherwise the advantages of SDHC seem to make the 151 a better choice than a 171 for the vast majority of people.

I bring it back to a similar assessment. It is driven by budget and the rest of what you are trying to do. If you have the budget, the DVCPRO HD will net you a better result in that it doesn't have to make it to 21 mbs and go long GOP. This is useful where there is a lot of motion and details in motion. It is also helpful when you are matching with the larger cameras, the texture of the codec will be similar. However, not everyone has the jingle in the pocket for P2 and DVCPRO HD, and the HMC150 does a very good job of making images that will stand up to any other low end codec and frankly will do a better job, as it is a great codec in its ability to smartly distribute its resources where needed. It is not always a 8X8 block, it is variable, going as small as 4X4 and as large as 16X16. As mentioned in a previous post, AVCHD is to HD what DV was to SD. The HMC150 is an awesome little camera and its implementation at 21Mbs, is unparalelled in the Long GOP world(within its price point striking diistance).

Best,

Jan

Jan_Crittenden
12-02-2008, 07:59 AM
email 2

"Yes, please do reproduce my reply to you if you wish. I’ve looked again at that thread since sending you the reply, and note that quite a lot extra has been subsequently said, and we seem now to have a Panasonic marketing person contradicting a Panasonic White Paper!
David"


Well since I wrote that White Paper I certainly know what it says. And what I am saying does not contradict what is explained in that paper.

Best to all,

Jan

Barry_Green
12-02-2008, 09:52 AM
Would then an MPEG-2 long-GOP at 100 megabits have more info than the DVCPRO HD intraframe?
Probably yes, provided that other factors are kept the same (such as recording frame size and color ratio).

Meaning, if you compared 4:2:0 MPEG-2 long-GOP against 4:2:2 intraframe, obviously the intraframe would have more info, all other things being equal. But the only implementation of 100mbps MPEG-2 that I know of is not exactly equal; it's 100mbps, and 4:2:2, but it's also full-raster 1920x1080 (as opposed to DVCPRO-HD's pre-filter of 1440x1080 PAL, or 1280x1080 NTSC).

So in a case like that, it'd be fair to say that yes the 100 megabit long-GOP codec would have more info than the 100 megabit intraframe, assuming (and it seems a reasonable assumption) that the nature of the long-GOP efficiency will overcome the additional bandwidth of the 33% to 50% larger recorded frame size.

BobDiaz
12-02-2008, 09:57 AM
Originally Posted by BobDiaz http://www.dvxuser.com/V6/images/buttons/viewpost.gif (http://www.dvxuser.com/V6/showthread.php?p=1483550#post1483550)
Correct me if I'm wrong about this, but I'm not aware of any HDTV camera for under $10,000 (USA price) that captures a full 4:2:2 color. Even the Sony EX1 & EX-3 are 4:2:0 color. Both the Canon A1 and H1 are 4:2:0 color.


Bob, you are wrong.

Best,

Jan

Let me re-state this..

Correct me if I'm wrong about this, but I'm not aware of any (Non-Panasonic) HDTV camera for under $10,000 (USA list price) that comes with a lens, that captures a full 4:2:2 color.

Did I miss something at NAB and DV Expo?

If yes, who makes it and what is the model?

Bob Diaz

trez
12-02-2008, 12:56 PM
This was my analogy to a container half empty. (If anyone disputes that, ask them what happens when confronted by a blue lines on black test chart, such that only the blue pixels produce an output. It CAN’T resolve better than 540 vertically.)

I'd been the advocate of this way of thinking for some time. Until yesterday - when I realized I'd been wrong.
Sure it's only blue sensor that is sensitive to blue color. It's true that blue sensor can't resolve more than 540 vertically. But having pure blue is rare in the real world.
Now, let's take cyan into account. Various cyan grades are much more common - they all consist of both green and blue. It is a MIX - just like luma - the only difference is that luma consists of all three - green, blue and red with proper weigths. So the cyan is the MIX of the two sensors' outputs - green and blue. Cyan is not resolved by blue sensor - it is resolved by both blue and green. Since green is half-pixel shifted against blue the spacial offset technology is capable of resolving more cyan-colored samples! Pixel-shifting works for chroma in exactly the same way it works for luma - because most colors are a MIX of various basic colors, just like luma, which is a MIX (weighted average) of all three basic colors.

BobDiaz
12-02-2008, 02:03 PM
If I had a bit more time I could produce screen grabs to show everyone, BUT I do have a class starting in about an hour and I do need to get ready for it....

If I take any photo of things, there is almost never a 100% pure red or 100% pure blue or 100% pure green in the colors. You can prove this with Photoshop, just take any photo and grab a color with the Eyedropper, and look at the RGB values for that color. Most of the time, it will have a mixture of Red, Blue, and Green. (Even that bright red car isn't 100% pure red.)

I can't do it now, but tonight, if I don't forget, I'll post some screen grabs showing this...


Bob Diaz

studio1972
12-02-2008, 03:54 PM
I'd been the advocate of this way of thinking for some time. Until yesterday - when I realized I'd been wrong.
Sure it's only blue sensor that is sensitive to blue color. It's true that blue sensor can't resolve more than 540 vertically. But having pure blue is rare in the real world.
Now, let's take cyan into account. Various cyan grades are much more common - they all consist of both green and blue. It is a MIX - just like luma - the only difference is that luma consists of all three - green, blue and red with proper weigths. So the cyan is the MIX of the two sensors' outputs - green and blue. Cyan is not resolved by blue sensor - it is resolved by both blue and green. Since green is half-pixel shifted against blue the spacial offset technology is capable of resolving more cyan-colored samples! Pixel-shifting works for chroma in exactly the same way it works for luma - because most colors are a MIX of various basic colors, just like luma, which is a MIX (weighted average) of all three basic colors.

I'm not an expert on this, but wouldn't the effect you are talking about result in either green or blue fringes along the edge on a cyan object. Does the camera have some clever algorithm to prevent this from happening?

BobDiaz
12-02-2008, 08:28 PM
OK, here's a look at some the the colors I've grabbed from Photoshop. I chose a shot I took at the LA Auto show with bright colors on the cars. Now, I chose to grab a lot of the darker parts of the colors, because the RGB values are always lower.

Here's the main shot:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture10.png

The dark blue of the headlamp still has some Green and Red in it:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture11.png

The dark green of the headlamp has some red and blue in it:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture12.png

The red of the headlamp has some blue and green in it:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture13.png

The red car paint (see the red car to the left) has some blue and green in it:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture15.png

The blue in my son's shirt has some green and blue in it:
http://i46.photobucket.com/albums/f103/KQ6WQ/Picture16.png

Those who have Photoshop can try it with this photo or any other photo, it's hard to find any normal color that is just 100% pure red or 100% pure green or 100% pure blue.


Bob Diaz

Bounce
12-03-2008, 04:14 AM
From David Heath:



"I’ve had another look at the thread you initially linked to, and it’s a bit frustrating to see some of the points I raised pecked over, given meanings and nuances that were never intended, and be unable to respond. Yes, it’s obviously true that total blue virtually never occurs in nature (let alone as perfect blue/black lines J), but that misses the point – it was intended as a hypothetical extreme example to underline a technical point.

Maybe the most relevant points are raised in posts 47 by trez, and post 49 by studio1972. I’ll try one last time to explain, and would be happy for you to post this e-mail on my behalf. Trez very sensibly brings up the question of “what about cyan?” so let’s use this as the basis of an example. It’s limiting not to be able to use diagrams, and the inevitable problem is knowing how far to simplify reality for clarity without losing too much accuracy, but I’ll do my best.

Imagine first a camera with 960x540 sensors, and no pixel shift. The active area is each is a square in the actual pixel site itself, and all the red, green, blue sites are aligned. Now imagine a horizontal line of light one pixel wide moves up the sensor, and to use Trez’s example it’s a cyan line. It should be obvious that it’s vertical position can’t be defined to any closer than 1/540th of frame height. Also, it must affect the green and blue sensors equally, so there can never be any ambiguity about what colour it is – cyan.

Now let’s introduce a half pixel pixel shift vertically, moving the blue pixels up by 1/1080 of the frame height to sit mainly directly over the dead spaces between photosites of the green sensor, and let’s imagine the same horizontal cyan line moving up.

Two things will immediately be different. Its position can be defined more precisely than 1/540 of frame height, and as it moves it will change from mainly illuminating blue sensors, to mainly green, to mainly blue etc. No longer is the colour unambiguous – it depends at what place we do the measurement. Do the same theoretical analysis with ANY colour horizontal line, and the overall conclusion is the same.

Which brings us on to the point made by studio1972 – “…wouldn't the effect you are talking about result in either green or blue fringes along the edge on a cyan object. Does the camera have some clever algorithm to prevent this from happening?”

And the answer to both those questions is a resounding “YES”, the latter a comparable process to De-Bayering with a Bayer sensor, and why the DSP in these cameras runs at 1920x1080 – it makes the maths easier.


To use video engineering terminology rather than mathematical, then in this case pixel shifting has undeniably improved the vertical luminance resolution (you know more accurately where the line is), but with no matching increase in chrominance resolution, which is the point I’ve been making all along. Pixel shifting works for luminance (resolution), not for chrominance. In the case we’re talking about here chrominance res sticks at 540 lines max.

Practically, implementations are far, far more involved than that, the theory gets much more complicated and much depends on individual chips, differing algorithms, to say nothing of lens performance. I fully realize that single pixel horizontal lines are as rare in real life as pure blue colours, but they are classroom devices to try and help people understand fundamental theory. We could go into matrixing aspects, or aliasing issues in a lab, but that wouldn’t help understand basic principles.

I’m not sure why all this seems so controversial, I’ve already said that I consider pixel shift a sound technique built on principles related to the physiology of the eye. I do think two things are being widely misunderstood. Firstly that luminance resolution and chrominance resolution are two separate factors, and secondly that this whole 4:x:x notation is only applicable to recording systems, not cameras as a whole.

Practically and in the real world, the question is how much it all matters? Overall, I’m the first to say “not that much”. But let’s remember the quote from me that was originally used to start this thread was initially a response to someone who asked about using an external recorder (CV XDR) with a 151, and was asking if the 422 aspect of the XDR would give a better result. To stop him wasting money, I replied I didn’t think so (though the lower overall compression may help), as the internal recording system of the 151 and it’s 1080 420 nature was already suited to the front end. I still think that’s true, I’ll leave everybody else to make up their own minds."

Jan_Crittenden
12-03-2008, 04:22 AM
[/I]
Correct me if I'm wrong about this, but I'm not aware of any (Non-Panasonic) HDTV camera for under $10,000 (USA list price) that comes with a lens, that captures a full 4:2:2 color.
If yes, who makes it and what is the model?
Bob Diaz

Okay, so any non-Panasonic, there is none and in fact you can raise that $ sign to $25,000 and it still wouldn't get you 4:2:2 in any make other than Panasonic. Within Panasonic yes, starting with the AG-HVX200A and the AG-HPX170. They record in full 4:2:2 called DVCPRO HD.

Best,

Jan

David Jimerson
12-03-2008, 07:16 AM
From David Heath:



"I’ve had another look at the thread you initially linked to, and it’s a bit frustrating to see some of the points I raised pecked over, given meanings and nuances that were never intended, and be unable to respond.

Why doesn't he just register and post? This third-party posting is awkward and should stop.

BobDiaz
12-03-2008, 11:08 AM
From David Heath:
...
Practically and in the real world, the question is how much it all matters? Overall, I’m the first to say “not that much”. But let’s remember the quote from me that was originally used to start this thread was initially a response to someone who asked about using an external recorder (CV XDR) with a 151, and was asking if the 422 aspect of the XDR would give a better result. To stop him wasting money, I replied I didn’t think so (though the lower overall compression may help), as the internal recording system of the 151 and it’s 1080 420 nature was already suited to the front end. I still think that’s true, I’ll leave everybody else to make up their own minds."

I wish David would sign up for the forum and post directly, but overall, we seem to be talking about something that is so slight, it's unlikely to matter. I get the feeling that some may believe that the edges of objects land exactly on pixel boundaries and transitional pixels do not exist; thus we could have some red+blue/green edge.

Several years back, Barry Green posted some images from the HVX-200 in the different modes. One of the images was DVCPRO50 (720x480). Now, even a 960x540 images sensor has more than enough pixels to resolve that, but when I blew the image up to 10x size, it shows that the edges are not as exact as you might expect:

http://i46.photobucket.com/albums/f103/KQ6WQ/DV50-Green_Zoom.jpg

So even if we had a 100% pure green background with a 100% pure (blue, red, or blue+red) line, the line almost never lies directly between the pixels. Thus, we will have transitional pixels between the boundaries of the background and the line.


I could get very technical here and go into sampling theory and why things work they way they do, but a simpler approach is to point to the HVX-200 section of this forum. Many of the users there also work with the higher end 2/3" cameras. If the 960x540 image sensors posed a real problem, we would have heard a massive outcry from the HVX-200 users years ago.

The perfect video camera does not exist, but given how popular the HVX-200 is, it shows that the technology does work and it works at a high enough quality level to get a loyal and happy group of users.


Bob Diaz

Justyn
12-03-2008, 01:24 PM
Bob.. always a good calming voice. Why does this rear up all the time. Who the hell has the time to care about such things. Get a cam.. and get to work with whatever tool is best for you. Shoot, some people are still buying DVXs..

David Saraceno
12-03-2008, 01:55 PM
I also wonder why Mr. Heath just doesn't register and post.

It's become a scene out of "Wizard of Oz."

trez
12-03-2008, 03:36 PM
Who the hell has the time to care about such things.

People planning their purchase. People confused with different opinions looking for real explanations. People wanting to know how things work. Your experience could really help them, Justyn.


Get a cam.. and get to work with whatever tool is best for you.

This won't help those people much. Someone who's buying the camera wants to first learn which tool is best for him and then get it. Also, you will never learn why spacial offset increases chroma resolution by running with your camera and shooting. To me the knowledge is precious and it seems there's bunch of people who think the same way.

It is ridiculous to me that some people complain about why this topic comes over and over again. There reason is we haven't had a good explanation yet.

Hidef1080
12-03-2008, 05:25 PM
Here's my two cents on this – as if anyone asked...


There is so much back and forth one the subject that after about the 3 page it becomes white noise.
We should research cameras and software before we use them but there are things that people won't/don't think about until after the fact and those people will 9 times out of 10 go online to known forums for insight or a point in the right direction.


At some point there has to be something conclusive or it will remain white noise.
Until then most people will fall back on “The images look good to me” and rightfully so.


People like Bob and Justyn keep it “real” so to speak and for that I say “thanks”.:thumbup:


Really, before I came to this site I thought little red, green and blue elves lived in my cameras painting pictures on my sd cards.:beer:

BobDiaz
12-03-2008, 09:27 PM
Just a bit of history, the debate of the 4:2:2 of the HVX-200 came up in early 2006 and it did get a bit ugly. I recall one person posted an insult to someone and became banned from the forum.

The whole thing ended up with the bottom line, "Look at the image that comes from the camera!!!" By then, there were enough people using the camera who could stand up and say it really looks good and had the results of green screen & blue screen tests to prove that it could be done without problems.

IF the image from the HVX-200 had sucked rocks, we would have heard about it long ago. Today, we have the same system in the HVX-200a, the HPX-170, and the HMC-150. The CCDs are improved, but from what I have seen, the improved CCDs have helped reduce the noise in the low light footage. So, the change is for the better.

I don't know of any image tests comparing the HMC-150 to the HPX-170; If anyone knows of such a test, please post the link to the results. Both cameras use the same lens and CCDs, so my own guess is that any difference is going to be small, but it would be nice to see proof one way or another.


Bob Diaz

Jan_Crittenden
12-04-2008, 02:58 AM
And I second Bob's synopsis, the 150 and the 170 look very similar, the recording codecs look a little different and from there it becomes application and/or budget driven.

All the best,

Jan

David Saraceno
12-04-2008, 01:16 PM
We have both.

What tests does anyone want run?

And I don't do charts.

BobDiaz
12-04-2008, 07:33 PM
We have both.

What tests does anyone want run?

And I don't do charts.

I'd love to see shots of the exact same thing. Say an outdoor shot with lots of colors and some detail in it. Assume the 2 cameras are set side by side.

HMC-150: at 720/24p and 1080/25p

HPX-170: at 720/24p and 1080/24p

Even just still photos would be nice to compare the differences or lack of difference...

Your comments on the videos would be nice... Do you see much of a difference in color or image quality.


Thanks,


Bob Diaz

Barry_Green
12-04-2008, 08:39 PM
Obviously an identically-lit, identically-shot greenscreen shot would be of interest. 1080/24pA on the HPX170 please. :)

David Saraceno
12-05-2008, 08:28 AM
Keyed or just shot?

Screen grabs or footage?

If footage, what codec?

Barry_Green
12-05-2008, 08:35 AM
Footage in the original codec, not keyed. People want raw shots to play with for themselves. So the original m2ts directory structure, and the original MXF directory structure, would be ideal.

BobDiaz
12-05-2008, 09:21 AM
A video is best, BUT for the sake of those who may not have the software to see the raw footage, still images would be helpful too.


Bob Diaz

Spartacus
12-05-2008, 10:08 AM
A thumb up for you efforts David, thanks!

David Saraceno
12-05-2008, 11:34 AM
See what I can do.

How about five seconds or so with each?

And someone to host the upload?

Justyn
12-06-2008, 08:54 AM
David... I can host that stuff for ya. Just hit me up. cheers


j

David Saraceno
12-06-2008, 09:16 AM
I'm going to set up our screen this weekend hopefully, and make certain my well paid talent will agreed to be videoed.

Heck, I married her, so she better.

SPrimeau
12-06-2008, 11:17 PM
Raw sample footage of both cameras. This is what I was looking for.

This will be really welcome.

Thanks in advance!

trez
12-07-2008, 05:59 AM
Raw footage AND stills, please...
Many people (including me) have had problems importing the AVCHD footage downloaded over the Internet to FCP even though the directory structure was preserved.
I had to transcode it first, which would make comparison unfair.
If you post the stills, let us know how they were created (which player).

I greatly appreciate your effort! Thanks

rtmst3k
12-14-2008, 11:36 PM
Hi All,

If the HMC-150 has the same "front-end" as the 170, I would assume that, at some point, it has access to a broader color space before a 4:2:0 image is recorded, and I was wondering if the HMC 150 HDMI port can output uncompressed, 4:2:2 ? If this is possible, it would seem like it might be possible to output the same image from the HMC-150 HDMI as is possible from the 170 HD-SDI?

And if the HDMI port outputs this uncompressed 4:2:2, are there HDMI recorders (with large hard disks) that could capture this (assuming these recorders are "somewhat" portable) ?

I'm new to this technology, so please forgive any ignorance or bonehead questions :)

Thanks!!
Randy

Hidef1080
12-15-2008, 04:35 AM
Hi All,

If the HMC-150 has the same "front-end" as the 170, I would assume that, at some point, it has access to a broader color space before a 4:2:0 image is recorded, and I was wondering if the HMC 150 HDMI port can output uncompressed, 4:2:2 ? If this is possible, it would seem like it might be possible to output the same image from the HMC-150 HDMI as is possible from the 170 HD-SDI?

And if the HDMI port outputs this uncompressed 4:2:2, are there HDMI recorders (with large hard disks) that could capture this (assuming these recorders are "somewhat" portable) ?

I'm new to this technology, so please forgive any ignorance or bonehead questions :)

Thanks!!
Randy

I have not used one yet however I am pretty sure that there are HDMI conversion boxes that will allow you to record images from the camera bypassing the AVCHD codec, giving you 4:2:2.
Recording the images to your hard drive via your NLE or some other software.

If I am wrong on this I'm sure someone here will point us in the right direction.

Barry_Green
12-15-2008, 06:32 AM
Yes, theoretically the HDMI output should be comparable (if not identical) to the HPX170's HD-SDI output. And yes there are or will be HDMI recorders; isn't the NanoFlash an HDMI recorder? You can output to a blackmagic intensity card through HDMI too.

Hidef1080
12-15-2008, 07:35 AM
Yes, theoretically the HDMI output should be comparable (if not identical) to the HPX170's HD-SDI output. And yes there are or will be HDMI recorders; isn't the NanoFlash an HDMI recorder? You can output to a blackmagic intensity card through HDMI too.

Nano Flash Recorder
http://www.convergent-design.com/downloads/nanoFlash/nanoFlash%20Brochure.pdf

Blackmagic HDMI to SDI
http://www.blackmagic-design.com/products/miniconverters/

trez
12-15-2008, 08:41 AM
If the HMC-150 has the same "front-end" as the 170, I would assume that, at some point, it has access to a broader color space before a 4:2:0 image is recorded, and I was wondering if the HMC 150 HDMI port can output uncompressed, 4:2:2 ? If this is possible, it would seem like it might be possible to output the same image from the HMC-150 HDMI as is possible from the 170 HD-SDI?


I haven't been able to get clear answer to this question so far. Panasonic says both cameras have the same lens, sensors and processing chip. They don't say anything about how the analog signal from the CCDs is processed.
Sure, the sample rate at the CCD is the same (and depends on luma resolution, which is usually higher than sensor native resolution) but what happens with this digital signal next?
Does 150 extract 4:2:2 chroma space first (and outputs such via its HDMI) and then downsamples it to 4:2:0 before AVCHD encoding?
Or rather the opposite is true? Panasonic might limit the performance of 150 on purpose (to not compete with 170) - in this case the 150 would get 4:2:0 chroma from combined pixel-shifted sensors and only upscale it for 4:2:2 HDMI output?

rtmst3k
12-15-2008, 01:50 PM
Nano Flash Recorder
http://www.convergent-design.com/downloads/nanoFlash/nanoFlash%20Brochure.pdf

Blackmagic HDMI to SDI
http://www.blackmagic-design.com/products/miniconverters/


So it sounds like you can record the output from the HMC-150 HDMI port, but you need TWO devices - an HDMI-->HD-SDI converter box, then you pick up an HD-SDI recorder like the nanoflash (records to SDHC cards).

Hope someone from Panasonic will consider making a recorder with an HDMI port onboard so we could have a "one box" solution for this...

Thanks,
Randy

Hidef1080
12-15-2008, 02:01 PM
So it sounds like you can record the output from the HMC-150 HDMI port, but you need TWO devices - an HDMI-->HD-SDI converter box, then you pick up an HD-SDI recorder like the nanoflash (records to SDHC cards).

Hope someone from Panasonic will consider making a recorder with an HDMI port onboard so we could have a "one box" solution for this...

Thanks,
Randy

I'm pretty sure you can record to your hard drive using Premiere Pro if your computer / laptop has an SDI port.
So I'm thinking HDMI-->HD-SDI-->NLE.

Again I have not done this myself but I'm almost 100% sure that it is an option.

http://livedocs.adobe.com/en_US/PremierePro/3.0/help.html?content=WS900278C5-C826-45dd-97B5-2AC27D6CC3CB.html

rtmst3k
12-16-2008, 09:04 PM
I'm pretty sure you can record to your hard drive using Premiere Pro if your computer / laptop has an SDI port.
So I'm thinking HDMI-->HD-SDI-->NLE.

Again I have not done this myself but I'm almost 100% sure that it is an option.

http://livedocs.adobe.com/en_US/PremierePro/3.0/help.html?content=WS900278C5-C826-45dd-97B5-2AC27D6CC3CB.html

Yes, that would probably work too, but it's still a 2-box solution.

I can think of two possible 1-box solutions:

1. A nanoflash-type product that has "HDMI-in" instead of "HD-SDI in". Recording
to a dedicated onboard disk.

OR

2. A laptop or a small desktop with some type of HDMI "capture" capability, but
the path from HDMI to disk would have to extremely fast assuming it's
uncompressed HD coming over the HDMI? Isn't uncompressed HD something
like 270 megabits/second?

Again, I haven't spent a whole lot of time yet researching these alternatives. I'm just getting started with all this research...

Thanks everyone!!
Randy

shrigg
12-17-2008, 10:37 AM
1. A nanoflash-type product that has "HDMI-in" instead of "HD-SDI in". Recording
to a dedicated onboard disk.

Doesn't exist, at least not yet... The other thing to consider is that these recorders cost more than an HMC-150 (nanoflash (http://convergent-design.com/downloads/nanoFlash/nanoFlash%20Brochure.pdf) projected price $3495+$395 for nanoconnect (http://convergent-design.com/CD_Products_nanoConnect.htm))


2. A laptop or a small desktop with some type of HDMI "capture" capability

Blackmagic Intensity (http://blackmagic-design.com/products/intensity/) which pretty much limits you to a desktop + SATA drives unless you have a macbook pro and are willing to void your warranty (http://www.dvinfo.net/conf/hdv2-canon-vixia-hv30-hv20-hv10/126256-blackmagic-intensity-macbook-pro-portable.html).
I wanna do this but I'm afraid.... :Drogar-Love(DBG):

jamesc
04-06-2009, 05:56 PM
I'm going to set up our screen this weekend hopefully, and make certain my well paid talent will agreed to be videoed.

Heck, I married her, so she better.

hey David, did you ever get a chance to capture the footage? I'm still debating between the HMC150 and the HPX170 and would love to get to test out keying on both.

Mike Schell
04-06-2009, 07:58 PM
So it sounds like you can record the output from the HMC-150 HDMI port, but you need TWO devices - an HDMI-->HD-SDI converter box, then you pick up an HD-SDI recorder like the nanoflash (records to SDHC cards).

Hope someone from Panasonic will consider making a recorder with an HDMI port onboard so we could have a "one box" solution for this...

Thanks,
Randy

Hi Randy-
The nanoFlash has both HD/SD-SDI I/O and HDMI I/O, so you can connect to either the HPX-170 or the HMC-150.

The nanoFlash records to Compact Flash cards, not SDHC. SDHC cards do not have sufficient bandwidth for our 100/160 Mbps CODEC. We are now recommending the Kingston 32GB 133X Compact Flash, which has sufficient performace for data-rates up to 140 Mbps and costs a whopping $60 at B&H.

We plan to debut the nanoFlash at NAB and begin shipments in late May.

Best Regards-
Mike Schell
Convergent Design

dcloud
04-07-2009, 01:50 AM
Doesn't exist, at least not yet... The other thing to consider is that these recorders cost more than an HMC-150 (nanoflash (http://convergent-design.com/downloads/nanoFlash/nanoFlash%20Brochure.pdf) projected price $3495+$395 for nanoconnect (http://convergent-design.com/CD_Products_nanoConnect.htm))

this pretty much blurs the line between hmc150 and hpx170 huh?

if the gh1 outputs live hdmi this would be an awesome combination

MDrake
06-13-2009, 02:31 AM
hey David, did you ever get a chance to capture the footage? I'm still debating between the HMC150 and the HPX170 and would love to get to test out keying on both.

I agree, it would be awesome to see a comparison of both cameras!