Panasonic 4K GH4 possibly coming next Spring...

[combatentropy]

Panasonic is going to place 200Mbps/100Mbps pro line codecs (that can't be displayed on a Panasonic consumer HDTV) in a consumer camera? Nonsense.

4k in a consumer camera? Sure, especially if Panasonic is trying to sell 4k televisions. But it sure as heck ain't gonna be a 200Mbps, 10 bit, 4:2:2 codec.

Do you know how old 4:2:0 (or 4:1:1) video is? At least 18 years. At the pace of technology, 10-bit 4:2:2 should have come to consumers years ago but is artificially held up by manufacturers' market segmentation. Market segmentation is not something to encourage, at least not by consumers. It's a trick companies use to inflate profits. It's why there were once seven versions of Windows out at the same time (Starter, Home, Professional, Enterprise, etc.). It's why easy-to-implement features are glaringly absent from SLRs.

At some point in even the cheapest HD video camera's signal chain is something like 1.5 Gbps 1920x1080 4:4:4 video. Maybe more. It is then pared down to whatever the media format is. The higher-end formats are subtracting less, compressing less, doing less, than the cheaper formats.

Even if you are cynical and doubt that a manufacturer will up the ante this year, why air it, encouraging them and discouraging us? Do you tell the car dealer, "You really should charge me more"?

 

[atarijedi]

At some point in even the cheapest HD video camera's signal chain is something like 1.5 Gbps 1920x1080 4:4:4 video. Maybe more. It is then pared down to whatever the media format is. The higher-end formats are subtracting less, compressing less, doing less, than the cheaper formats.

Not necessarily.

It comes off the sensor as raw value, it is demosaiced and at this point it is essentially an RGB colourspace, and then it might be chroma subsampled to 444, or to 422, or to 420, or whatever. For instance, if they are going to encode it as AVC Intra 100, they won't subsample to 444, they will skip right to 422, for AVC Intra 50 or AVCHD, they will skip right to 420. Or they may use a sampling method which doesn't involve Y'Cb'Cr chroma subsampling at all, like DPX 10bit RGB.

 

[Danielvilliers]

Perhaps people should look at it the other way round. Why would Pana put a 4k camera with high bit rate and 10 bit in a GH camera... because it is a gh camera. It won't cannibalise its own traditional camcorder/broadcast camera line. People that will buy varicam camera won't go out to buy this Ag-gh hybrid even if it is 4k etc. This rumoured camera might also be a good test bed for its future 4k camcorders.

It might also be internal changes within Panasonic where the two Lumix and video division share more tech etc. It might even be the Lumix division taking more power of decision for two reasons. They have been so much more innovative than the video division even if they are much younger team and I am sure that the Lumix line brings much more money to Pana now than its video division.

The last thing I would add is that pixel size has no more importance to think about overall sensitivity and dynamic range. For example at pixel level, my friend Nikon 12 megapixel d700 is a bit better than my D800. But my D800 has 36 megapixel of them so when you compare them at output level (same size photo) the D800 is better. If it really samples the 16 megapixel of the sensor and downscales it to 4k or 2k, this camera overall sensitivity could be very good. My D7100 that does some pixel binning and not just line skipping, rivals the 5d mark3 in low light and beats my D800 and I am sure many other full frame camera in video.

 

[maarek]

Do you know how old 4:2:0 (or 4:1:1) video is? At least 18 years. At the pace of technology, 10-bit 4:2:2 should have come to consumers years ago but is artificially held up by manufacturers' market segmentation. Market segmentation is not something to encourage, at least not by consumers. It's a trick companies use to inflate profits.

This is a complete misunderstanding of business. There are clients who do not want video in their DSLR's (surprisingly many) and would not pay a dime for them. Why put any resources there?

Segmentation happens because people have different wants and needs.

Saying that 10bit 4:2:2 could've been recorded onto minidv tapes 10 years ago is unbelievable hogwash.

It baffles me how some people think they are entitled to all kinds off stuff.

 

[Gillvane]

Do you know how old 4:2:0 (or 4:1:1) video is? At least 18 years. At the pace of technology, 10-bit 4:2:2 should have come to consumers years ago but is artificially held up by manufacturers' market segmentation. Market segmentation is not something to encourage, at least not by consumers. It's a trick companies use to inflate profits. It's why there were once seven versions of Windows out at the same time (Starter, Home, Professional, Enterprise, etc.). It's why easy-to-implement features are glaringly absent from SLRs.

At some point in even the cheapest HD video camera's signal chain is something like 1.5 Gbps 1920x1080 4:4:4 video. Maybe more. It is then pared down to whatever the media format is. The higher-end formats are subtracting less, compressing less, doing less, than the cheaper formats.

Even if you are cynical and doubt that a manufacturer will up the ante this year, why air it, encouraging them and discouraging us? Do you tell the car dealer, "You really should charge me more"?

I don't believe you are using the term "market segmentation" correctly. This is what it actually means:

http://businesscasestudies.co.uk/business-theory/marketing/segmentation.html#axzz2iplx7VtQ

The only time it would be harmful to consumers is if there was a monopoly and then the cause of the harm would actually be the monopoly. The market segmentation would just be a way to take even more advantage of the monopoly. I don't think any company has a monopoly on cameras.

Generally market segmentation is good for consumers. It just means a business is catering to the needs of different groups. The example in the link is selling sandwiches. A vendor might sell vegetarian and non-vegetarian sandwiches to cater to both vegetarians and non-vegetarians. That would be market segmentation. I would not say this is "bad for consumers" and that we should not encourage it.

I expect the car dealer to charge me as much as they can. Since the car dealer doesn't have a monopoly if they are charging to much I will walk across the street and purchase a different make and model of car. If Sony is charging to much I will walk across the street to Canon. If both are charging to much I expect someone to take advantage of that and offer a new camera at a lower price like the Black Magic Cinema Camera.

 

[Vultch]

Computers, Automobiles, mobile phones, each year we see new features and improved specs.
Why shouldn't video camera be offering 4:2:2 as standard, 4:1:1 or 4:2:0 is getting a little old and I cant imagine the cost of producing the circuitry is any more expensive.

 

[harddrive]

Isn't the difference in resolution mostly due to the different aspect ratio of video and stills though? Doing some rough math indicates the 16:9 portion would be pretty close to 4k.

Not really - 16:9 crop out of 4:3 sensor is 75%, so about 12 Mp active - it's about 8.3Mp (3840x2160) for nearly all the 4K s35 sensors. A lot fewer photosites in twice the area.

What do you base that statement on? I've heard that there's little improvement to be made on tiny cell phone sensors, but it seems like there are still big leaps being made on cinema sized sensors. For example Red's Dragon sensor which is massively more sensitive and with more dynamic range despite the photosite size being slightly smaller than their previous MX sensor.

Try http://www.clarkvision.com/articles/digital.sensor.performance.summary/ as a reference. He states that photon absorption efficiency is about 50% for modern digital cameras, and that relates to other things I've seen. Obviously, for a given f no and shutter speed, the total no of photons/photosite will depend heavily on the area of each. The max possible no of photons that can generate electrons will therefore heavily determine the dynamic range, for one.

But note that even if the figure could be increased to 100% (which it won't), that only represents a doubling or a theoretical max possible improvement of 1 stop.

 

[harddrive]

Do you know how old 4:2:0 (or 4:1:1) video is? At least 18 years.

Yes, but do you know how old 4:2:2 is? Considerably older, it dates from 1986 and was the subsampling used in the first digital component VTR standard - D1. See http://en.wikipedia.org/wiki/D1_SMPTE .

What is less appreciated is why 4:2:2 exists as a standard at all, and the whole subject of colour subsampling is one where maybe the most nonsense gets talked about in the whole field of video.

Back in the 1980's, when digital techniques were quite new to video, it was seen as desirable to reduce the data rate for obvious reasons. Exploiting the way the eye is less sensitive to chroma detail than luminance was an obvious follow on from the way that technique had already been used in analogue PAL, NTSC systems - hence less samples assigned to chroma than luminance. But 4:2:2 halves the horizontal samples, but not the vertical - have you ever wondered why? The answer is simply due to it being developed for interlace systems. Halving vertically is problematic then - done on a simple line by line basis it would mean only one field had any chroma information!

Move on nearly 10 years to the time you talk about (and DV) and the desire was there to create a system that was much, much cheaper than such as D1, but still give recording quality superior to the BetaSP of the time - even in consumer cameras. (Forget about the front end, for now.) Halving the chroma samples again was seen as necessary - but should it be done horizontally (again) or this time vertically?

Initially it was thought horizontally, again because of interlace reasons, and hence 4:1:1. That was chosen when digital VTs were seen largely as digital islands in an analogue world - but that didn't last long. It was rapidly seen that 4:2:0 was more suitable when signals would largely stay digital, and because DV and DVCAM PAL versions came later, that's why in PAL DV and DVCAM are both 4:2:0.

OK, if you're still awake and still reading, to come to the point. With the growing use of progressive video, 4:2:2 and 4:1:1 as systems lose much of the point. With progressive video, if you need to reduce data it's logical to treat the horizontal and vertical subsampling equally - to suit the way the eye behaves. Hence halving in each direction is most easily done - 4:2:0. If a lot of compositing is to be done, that may not be acceptable, but in that case why halve in one direction but not the other? There is a case now (with progressive systems) for 4:2:0, a case for 4:4:4. But 4:2:2 is a hangover from the days of interlace and doing D-A/A-D conversions on the inputs/outputs in the broadcast chain.

Unfortunately, all that historical logic has got largely lost and "4:2:2 good, 4:2:0 bad" has just become a mantra.

At the pace of technology, 10-bit 4:2:2 should have come to consumers years ago but is artificially held up by manufacturers' market segmentation.

Unfortunately, 10 bit 4:2:2 needs higher data rates than 8 bit 4:2:0 - so higher overall compression or more data to record to tape/card, whatever. That means both larger memories and even more significantly FASTER memories to handle the datarate. And until recently, THAT is what was the main determining feature - nothing artificial.

It's difficult for the manufacturers to keep pace though - AVC-HD was developed by Panasonic and Sony specifically as a way to get decentish quality onto cheap consumer memory, SDHC cards. But the first cameras are barely out when it's realised that the cards have now progressed such that somebody works out how to record XDCAM EX to such cards - so what's really the point of AVC-HD after that? It may offer slightly smaller file sizes - but is much more difficult to natively edit. It's real selling point was using low cost memory - but if you can now record XDCAM to that, why bother?

That said, memory costs have come down considerably in many ways and I do think we're seeing signs that products will increasingly come to take advantage. What about the Sony PXW-Z100 and it's use of XAVC via XQD?

At some point in even the cheapest HD video camera's signal chain is something like 1.5 Gbps 1920x1080 4:4:4 video. Maybe more. It is then pared down to whatever the media format is.

For a 3 chip 1920x1080 camera, that may be true in principle - but it's not true for the cheapest. That's especially so when we're thinking of still camera sensors which can clearly be seen on test charts not to give anything like 1920x1080 in video mode.

 

[harddrive]

Perhaps people should look at it the other way round.

It might even be the Lumix division taking more power of decision for two reasons. They have been so much more innovative than the video division even if they are much younger team and I am sure that the Lumix line brings much more money to Pana now than its video division.

Of everything written in this thread, that is maybe the most interesting thought. It's easy to think of such as Sony or Panasonic as one single entity, but in practice that's not the case. I can well believe more rivalry existing within a single company between divisions than between companies!

And it does seem to be Panasonic video that's been really having the problems. Maybe the point about "just think what it'll do to their existing higher price pro ranges...!" is met by they don't care - because they are a different division!?

 

[hscully]

This thread has been very informative. One thing I like about dedicated video cameras is that they have OLPFs. I think that the aliasing and moire involved in having a still camera shoot video is really horrible. I have a GH2 and I like it less than my AF100 because of that.

 

[Gabrobot]

Not really - 16:9 crop out of 4:3 sensor is 75%, so about 12 Mp active - it's about 8.3Mp (3840x2160) for nearly all the 4K s35 sensors. A lot fewer photosites in twice the area.

16mp for a 4:3 sensor comes out to something like 4620 x 3465, which means the 16:9 crop would be about 4620 x 2599...that seems like about the right amount of oversampling used in a lot of bayer sensor video cameras.

Try http://www.clarkvision.com/articles/digital.sensor.performance.summary/ as a reference. He states that photon absorption efficiency is about 50% for modern digital cameras, and that relates to other things I've seen. Obviously, for a given f no and shutter speed, the total no of photons/photosite will depend heavily on the area of each. The max possible no of photons that can generate electrons will therefore heavily determine the dynamic range, for one.

But note that even if the figure could be increased to 100% (which it won't), that only represents a doubling or a theoretical max possible improvement of 1 stop.

What's the real world problem here though? Even if it performs only as well as the GH2, that's still probably going to be better than the 4k BMCC. I don't care about the theory, I care about the images I can create.

 

[Barry_Green]

But 4:2:2 is a hangover from the days of interlace and doing D-A/A-D conversions on the inputs/outputs in the broadcast chain.

Unfortunately, all that historical logic has got largely lost and "4:2:2 good, 4:2:0 bad" has just become a mantra.

David, am I reading you right? Are you saying that you consider 4:2:2 archaic, a leftover, and not an improvement over 4:2:0 in progressive video?

I can't believe that's what I'm reading, but that's what it sounds like you're saying...

I won't disagree with you about the origins of 4:2:2 being that they wanted chroma on every field, as 4:2:0 in interlaced is a hideous nightmare and almost as bad an affront to reason as ... well, as interlaced video in the first place. And I will agree that 4:2:0 makes more sense in progressive than 4:1:1 would.

But to say 4:2:2 is just a mantra? 4:2:2 is literally twice the color resolution of 4:2:0. No it's not square, but I'll take 2x the color information (and the attendant increase in color accuracy that comes with it) over square pixels any day of the week. 4:2:2 is substantially better than 4:2:0, any time, anywhere, any place, any scenario, any circumstance.

There are some pretty good reasons why the best video engineers in the world set the highest broadcast standards at 4:2:2 instead of 4:2:0. Sure 4:4:4 would be better, but the increase in discernible image fidelity is pretty minor over 4:2:2, so it's generally not considered an efficient use of bandwidth to bother encoding 4:4:4 (i.e., you can usually get better results by allocating those bits elsewhere).

 

[harddrive]

David, am I reading you right? Are you saying that you consider 4:2:2 archaic, a leftover, and not an improvement over 4:2:0 in progressive video?

I can't believe that's what I'm reading, but that's what it sounds like you're saying....

No, that's not really what I am saying.

It is an improvement - it can't not be - but there has developed this "mantra" of "it must be 4:2:2!", and whilst that may be justified in some cases, it's much less so in others. What I'm trying to do is not say 4:2:2 is "archaic", just try and put some reason back into the whole subject. The original post I responded to seemed to imply a belief that 4:2:2 is more modern than the "old" 4:2:0, which is absolutely not the case - it predates it by over a decade.

And whilst 4:2:2 may have offered very significant gains over 4:2:0 in the days of SD interlace, the impact is less when we talk about HD and progressive. Not none, by any means, but it's nowhere near as significant as in SD interlace days.

It also seems to come across that the 4:2:2/4:2:0 subject has developed more significance in the NTSC world than the PAL world, and I believe that goes back to DV standards. Firstly, 4:2:0 (even for SD interlace) isn't too bad if used just for A-D coding/D-A decoding and display. It's when you get repeated analogue-digital-analogue recodes that you get the problems, and this dates from what I said about the early digital VTs being first envisaged as "digital islands" in an analogue world. In such case, 4:2:0 and interlace can lead to vertical chroma shifts in multi-generation, and this is precisely why the first NTSC standards for the DV family were 4:1:1.

But this situation didn't last very long. The writing was on the wall before the PAL specs for DV were finalised, when it was seen that the signal was more likely to remain digital through the chain, even to transmission, and to DVD - then about to take over from VHS. Hence DV, DVCAM (though not DVCPro) are 4:2:0 for PAL. The importance of this is that most final delivery codecs are 4:2:0 - DVB transmission, DVD and Blu-Ray. Whatever origination and editing are in, it's likely to be subsampled to 4:2:0 for the end product. So in PAL, it's fine - output is the same as an original DV input, at least as far as subsampling goes. (One chroma sample for every four luminance, and horizontal/vertical symmetry.) But in NTSC it's much nastier. The 4:1:1 subsampling means an original 1 in 4 chroma reduction horizontally, then a further culling of the chroma by a factor of 2 for the 4:2:0 compliant DVD. Nasty. Only one chroma sample for every EIGHT luminance.

With all NTSC acquisition formats either 4:1:1 or 4:2:2, this was largely why 4:2:2 got seen as such a "good thing" in the NTSC world. When subsampled for such as DVD coding, you got the 1 in 4 ratio - not 1 in 8. The real benefit of using a 4:2:2 system for acquisition was not so much that it was 4:2:2 per se - more that it WASN'T 4:1:1!

Move to the HD world and 4:1:1 (and 3:1:1) have gone away, I'm pleased to say. But a lot of the positive gut feeling about 4:2:2 that the above situation generated remained.

Come HD and the most common broadcast standard is 1080i/25 (1080i/30 in the 60Hz world) - so interlace. And that's why 4:2:2 is still considered the norm for normal broadcast work - and why I'm certainly not calling it "archaic". But with Blu-Ray and DVB transmission standards 4:2:0, and no 4:1:1 HD standard, the situation is radically changed from SD days. You say "the best video engineers in the world set the highest broadcast standards at 4:2:2 instead of 4:2:0 which, yes, is currently true for 1080i/25 acquisition and post - but not for transmission. UK DVB HD transmission is currently 1080i/25 at 4:2:0, and same for most of Europe.

But move to 4K (as this rumoured camera, with it's rumoured codec, is) and it's another step change. There is no 4K interlace standard anyway, and with the base resolution so high, then for 4K 4:2:0 the chroma resolution is still 1920x1080! Given the science of the human eye being less sensitive to chroma resolution than luminance, and what size of screen do you think will be necessary before 1920x1080 chroma starts becoming a significant limitation?! Likewise, most 4k cameras at the moment (even at the higher end) use a 3840x2160 Bayer sensor, which is only likely to give a chroma resolution of 1920x1080 anyway.

What I'm really trying to say is that in the days of SD, interlace, and especially in the NTSC world, there were very real and significant advantages to using 4:2:2 acquisition. Nowadays (and how much is dependent on many factors) there may still be advantages - but they are proportionally far less, especially in the progressive world. For the highest quality work and post production then higher chroma resolutions may indeed be desirable, but why reduce by a factor of two horizontally but not vertically? What I'm really trying to say is that once the interlace factor is lost, 4:2:2 starts to sit more uncomfortably between the stools of 4:2:0 and 4:4:4.

 

[Barry_Green]

There is reason in much of what you say, except when you start to point out transmission -- I have never let transmission dictate acquisition, and I don't think any reasonable person should. Transmission has always ruined signals, but that doesn't mean we should start acquiring on crappy low-res video systems just because the transmission is going to ruin it... and yes, I have heard that argument advanced, when the DSLR days were young and I was trying to point out just how horrid the moire problem was, and people would post threads saying "I saw moire on my HDTV last night, so it's everywhere and it's no big deal, so I'll just use my moire-filled DSLR and everything will be fine."

I'm sure you'd agree with me.

So the notion that transmission is 4:2:0 is irrelevant... just as irrelevant as that transmission in my country is in the creaky, inefficient, please-let-it-die MPEG2. I know that broadcast will destroy my image fidelity, I can't help that, but I can darn well ensure that my product looks good up until the point where it gets submitted for broadcast -- or DVD, or streaming, or Blu-ray or whatever else gets done with it. But I insist on higher standards on the front end, just like the BBC does -- acquisition standards are higher specifically because that's the only way you have a hope in Hades of your signal surviving the butchery of today's broadcast standards!

Okay, last things:

For the highest quality work and post production then higher chroma resolutions may indeed be desirable

Why only "may"? Under what imaginable circumstances would they be less desirable? I can't think of any.

but why reduce by a factor of two horizontally but not vertically?

Surely this is obvious, and was referenced in my last post -- because if you encoded 4:4:4, you'd be using up a lot of bandwidth for a minimal improvement in overall image quality. And you can't reduce by 1.5 pixels, it's got to be a full pixel or not at all. So if the question is: you can have 25% color resolution, or 50% color resolution, I know which my pick will be. And I'd give the same answer if it was 1x2 blocks instead of 2x1 blocks (i.e., recording full bandwidth chroma horizontally but only every other line). It has nothing to do with interlace anymore, it's just down to the simple math. More is better, and the attendant bandwidth increase is well worth the image quality and color fidelity increase.

What I'm really trying to say is that once the interlace factor is lost, 4:2:2 starts to sit more uncomfortably between the stools of 4:2:0 and 4:4:4.

Totally disagree. Benefit is benefit.

Look, if we get back to it, why did 4:2:0 come about anyway? Because DVD delivery and HDTV delivery wanted to squeeze down the absolute minimum bitrate. Period. That's the whole reason it came about. There wasn't any debate over "should we make it square chroma samples so it'll feel more artistically orthogonal" or whatever, it was "what can we hack and slash that won't make the image too awful?" And chroma resolution got the hack. But is this a good thing? Absolutely, unequivocally not. It is an absolute here that less is, indeed, less. It's not the worst compromise in the world, but in the same breath it is not something to aspire to. I don't know anyone who says "yay, my video is 4:2:0!" If we want something to aim towards, we should aim towards 4:4:4. 4:2:0 is a big step away from the compromised, bitrate-slashed, chroma-decimated 4:2:0 that has become commonplace.

When you get right down to it, 4:2:0 is not good, it's barely tolerable. It gets the job done for the minimal-bandwidth acquisition systems. But that's where its usefulness ends. 4:2:2 is vastly better, because it's a full doubling of color resolution. No, it's not square pixels, but square pixels are impossible -- you're either going to have 2x1 or 1x2, you can't have 1.4 x 1.4 pixels, so -- once you get over the notion that it's not square, but that you've got twice the resolution, how can that be seen as anything but a good thing?

In acquisition we should be pushing for more. 4:2:0 in interlace was a hideous abomination. 4:2:0 in progressive is fine for the compromise that it is, but it isn't ideal for compositing, keying, greenscreen, or significant post work. 4:2:2 is much better for all those tasks.

 

[harddrive]

Look, if we get back to it, why did 4:2:0 come about anyway? Because DVD delivery and HDTV delivery wanted to squeeze down the absolute minimum bitrate. Period. That's the whole reason it came about. There wasn't any debate over "should we make it square chroma samples so it'll feel more artistically orthogonal" or whatever, it was "what can we hack and slash that won't make the image too awful?" And chroma resolution got the hack. But is this a good thing? Absolutely, unequivocally not.

There is far more science to it than that. In an ideal world, it would be fantastic to transmit a full uncompressed HD-SDI signal to every home. Fantastic, even, to simply record the full uncompressed HD-SDI signal.

Practically, it's necessary to reduce bandwidth. The job of broadcast engineers is to do that in a way which most practically maximises quality *OVERALL* for any given allocated bandwidth - which must involve a series of compromises. The trick is to optimise the compromises to get the best overall result. And heck, even using an R,G,B system in the first place is a huge compromise in the first place, rather than accurately reproducing spectral colours. But - it's making use of the physics and biology of human vision. It's making use of the fact that the human vision system has it's own ways of reducing raw information to be manageable.

And as to why 4:2:0 - halving chroma resolution horizontally and vertically with respect to luminance - came about, it's not as random as you make out. It's making use of the well known fact that the human eye is far less sensitive to chroma resolution than luminance. And it makes sense to do it symmetrically - that also matches the human eye. This is nothing new, and the halving of chroma resolution here is less than was the case for analogue systems, both NTSC and PAL. The whole point is to lose information that the eye can't use - and losing chroma resolution is an efficient way of doing it. In the case of NTSC/PAL, it's done by a far greater factor than 2:1.

Many of the problems stem from the notion that "4:2:2" and "4:2:2 v 4:2:0" are absolute matters. The numbers are simply ratios, and whilst, yes, you may argue "more must be better" then how much better is far more variable than generally accepted. And that's down to matters of interlace or not, SD, HD... or 4K. I don't deny that for SD and interlace the benefits are large. But proportionally, they are far, far slighter for progressive and 4K. That's largely down to in the case of SD systems, the luminance resolution is lacking as far as the eyes perception is concerned. So even allowing for the lower perception of colour, you're starting from a lower baseline. (Quite apart from interlace issues.)

But move to HD - and especially 4K - and it's a different story. With 4:2:0 progressive and 1080, then the chrominance resolution alone is far higher than luminance for NTSC! A lot more so when you take Kell factor into account. With 4K and 4:2:0, chroma resolution is 1920x1080 - normal luminance for HD. And I'd argue that you'd have to have a very, very large screen indeed for the eye to notice a chroma resolution better than that - so why go to 4:2:2 at 4K?

I think there are actually two distinct cases. What's good for straightforward viewing, and what's needed for more advanced post work. I'd argue that 4:2:0 is fine for the former for 4K, and probably for normal HD as well, at least for progressive. For advanced post work, there may be an advantage to less subsampling - as then chroma resolution can affect eventual luminance resolution.

But the question is how much. As you say, the options realistically are 25%, 50% and 100%, and this is where we differ. You see 50% as the sweet spot - to me it falls between stools. I see 25% (4:2:0) as a good balance for "straightforward viewing", but if the work is so critical that more is needed you really need 100% (4:4:4) - at this level, 4:2:2 is neither one thing nor the other. Lacks the efficiency of 4:2:0, yet also the absolute quality of 4:4:4.

4:2:2 is vastly better, because it's a full doubling of color resolution. No, it's not square pixels, but square pixels are impossible -- you're either going to have 2x1 or 1x2, you can't have 1.4 x 1.4 pixels, so -- once you get over the notion that it's not square, but that you've got twice the resolution, how can that be seen as anything but a good thing?

Because I'll turn that question round. Assuming a progressive system - if the doubling of that colour resolution is so noticeable in the vertical axis, why is it not important on the horizontal axis? If you need that chroma resolution vertically, how can you get away without it horizontally? Why do you consider horizontal lines to have so much more significance than vertical? Human vision acts symmetrically - shouldn't the systems we use in broadcast and cinematography be the same? Hence I see the value of 4:2:0, I see the value of 4:4:4 - 4:2:2 in this scenario seems neither one nor the other, not a sweet spot.

Of course, my questions above get fully answered for interlace systems. They are asymmetrical by definition, 4:2:2 suits them perfectly, and are why it exists.

And that's why it's so wrong to just talk of "4:2:2" etc as an isolated concept. In some situations it's absolutely right, in others it's not. That's why I originally objected to the way it gets trotted out as a seeming mantra.

 

[harddrive]

What's the real world problem here though? Even if it performs only as well as the GH2, that's still probably going to be better than the 4k BMCC. I don't care about the theory, I care about the images I can create.

No problem at all. Go back to the original quotes and I was commenting about the view that new tech is coming along and it was speculated that it could lead to "massive improvements in sensitivity". There are physical reasons to see that as optimistic at the very least. It may be a bit better than current technology - but it won't be "massive".

 

[Barry_Green]

Clearly we are not going to agree. And that's okay, as I'm very, very, very comfortable in my view. 4:2:2 is better than 4:2:0, and there is no possible way to argue that it isn't. Which is why, I think, you insist on arguing against things that aren't part of the discussion. You want to talk about symmetry, as if the goal of symmetry is more important than having double the resolution. I dare say that you will find few to agree with that assumption. And even if you did find someone who would agree with that, that still wouldn't address the fundamental premise -- more resolution is better than less. It just is.

4:4:4 is better than 4:2:2. Nobody's arguing that. 4:2:2 is better than 4:2:0, empirically, fundamentally, evidentially, observably, scientifically, mathematically, and in all other ways. You seem to disagree. I can make one case for 4:2:0 being better than 4:2:2, and only one case -- if you were so starved for bitrate that you were making barely-viewable images (think HDV), then spending some of that precious bitrate on 4:2:2 would be the wrong decision, it's needed more to maintain a somewhat viewable image structure overall, and 4:2:0 would be the right choice in that case. But only in that case.

It isn't that 4:2:0 is awful, it's acceptable for most viewing purposes. But your fascination with the human eye is clouding your judgement and rendering your argument invalid in my opinion -- the digital acquisition systems are most definitely not making images for the human eye, they're making images that can compress well, be manipulated, be graded and keyed, and have realistic color reproduction. You can argue for 4:2:0 for transmission all you want, that's fine, I don't care about transmission, as I will never have control over transmission. But for acquisition, better is better. 4:2:2 should be considered the minimum for acquisition whenever practical or possible, and we shouldn't be arguing that "4:2:0 is just as good because of the ratio of rods vs cones in the human eye"... that's just so totally missing the point that it's hard to fathom why you're making that argument. Perhaps you're still using XDCAM EX and therefore want to assert that 4:2:0 is good enough? But if so, just come out and state it that way -- say "4:2:2 is better than 4:2:0, but my clients can't tell the difference and I don't do the type of work that needs 4:2:2, so I'm finding 4:2:0 plenty adequate." That'd be fine with me -- you'd be stating a personal preference. But that's not what your argument has been, your argument has been to denigrate the position of 4:2:2 in the world of acquisition, and I just can't agree with you at all on that.

 

[pulpfiction007]

I agree with getting the best acquisition you can afford on the front end, even though you may wind up losing some of the gains in the delivery format. I used to record audio in 16 bit, and I reasoned that it was good enough because the final delivery (CD) was only 16 bit. I couldn't have been more wrong. As soon as I moved to 24 bit there was a noticeable leap in quality. It comes down to "headroom" in the signal....anything that can deliver more of that for you is a good thing.

 

[Bassman2003]

I agree with Barry but admit do not know what the argument is about! More data is always better in aquisition... Audio is a good and bad example as it does improve a lot but the jump from 16 bit to 24 bit is massive on a numbers scale. 4:2:0 to 4:2:2 even to 4:4:4 is not on the same scale if they are all in 8 bit. A simplar leap is 8 bit color to 14bit RAW. Large difference in what you can do with the 14 bit RAW.

4:4:4 is just a huge memory commitment. As prices come down with time this will come more into play but 4:2:0 is the consumer format because of memory size. Consumers do not want large file sizes. Convenience trumps everything else in the "real world" - not the video geek world we live in. I had a client ask if I could e-mail the 4-hour lecture they gave...

With HD production, a 4:2:2 50mbps long GOP based upon the h.264 codec like the new Panasonic variant will cover most everything under high end cinema production imho. I have been very happy with the 4:2:2 mpeg-2 stuff and that is a less efficient codec.

 

[harddrive]

Clearly we are not going to agree. And that's okay, as I'm very, very, very comfortable in my view. 4:2:2 is better than 4:2:0, and there is no possible way to argue that it isn't.

Barry -I have NEVER argued anything other than 4:2:2 is indeed better than 4:2:0 - the question is rather by how much, and how does that vary with circumstance? And THERE is where we may differ. I'm sure you'll have no disagreement if I say that for SD and an interlace system, the answer is "a lot"? But move to HD and progressive and the benefit is much less. Go to 4K and progressive and the difference is less still. I said in the last post that "it's so wrong to just talk of "4:2:2" etc as an isolated concept" and I reiterate that now.

You want to talk about symmetry, as if the goal of symmetry is more important than having double the resolution. I dare say that you will find few to agree with that assumption.

The goal of symmetry here is to make the most of what you have. Equivalent in a way to saying that anything is only as strong as it's weakest link. And if you want to double the strength of something, there's little point in doubling the strength of part A, whilst leaving part B as before. It will still break at the same force, just always at part B.

And similar principle here. Assuming progressive, then if half chroma resolution is found unsatisfactory for whatever reason, then why double it vertically and not horizontally? Why only do half the job? If half resolution is unsatisfactory vertically, why is it fine horizontally? That's the significance of symmetry.

Please - would you or anybody else like to answer just that one question: "if half chroma resolution is found unsatisfactory for whatever reason, then why double it vertically and not horizontally?"

Take the symmetry away and it's a completely different story. And that's exactly what happens with an interlace system. And that's exactly why 4:2:2 was the first digital component subsampling scheme. It was developed when all television was interlace, and that is the main reason why it's so preferred still today - as most broadcast HD is still interlace. The asymmetry of 4:2:2 compensates for the asymmetry of interlace.

You say that 4:2:2 is double the {chroma} resolution of 4:2:0. Which it may indeed be - but all on one axis only. I suppose something is always better than nothing - but it's not the same, nowhere near as satisfactory, as a doubling of overall resolution whilst keeping symmetry. (Do I have to qualify with "in a progressive system" here?)

4:2:2 is better than 4:2:0, empirically, fundamentally, evidentially, observably, scientifically, mathematically, and in all other ways. You seem to disagree.

No, I don't disagree, have never said otherwise, and again, I can only say check back over everything posted in the last couple of threads to see that's true. It may always be better - but the difference is far more relevant in some cases than others, that's the point.

But your fascination with the human eye is clouding your judgement and rendering your argument invalid in my opinion -- the digital acquisition systems are most definitely not making images for the human eye, they're making images that can compress well, be manipulated, be graded and keyed, and have realistic color reproduction.

Come on, the human eye has everything to do with the situation!! Even in the most complicated and expensive process, the object is to end up with an image on a screen that represents the subject as accurately as possible - to the human eye. We're not making films or programmes for aliens or cats and dogs! Scientifically, it won't really be anything like the original scene in many ways - it won't have anything like the accurate full spectral response, it won't have the fluidity of continuous motion, and yes, the chrominance resolution will be less than luminance - but the point is that TO A HUMAN it will closely seem as the eye would see the real scene. (Try to watch "Planet Earth" with a pet dog and don't be surprised if it doesn't get excited by the animals on screen - it's more likely to just perceive a flickering light.)

But that's not what your argument has been, your argument has been to denigrate the position of 4:2:2 in the world of acquisition, and I just can't agree with you at all on that.

Barry, that is a misrepresentation of what I've been arguing. Maybe I haven't been clear enough, in which case please do me the favour of rereading the last couple of posts with an open mind if you really believe that.

Firstly, I've tried very hard to show why "4:2:2" is better in some circumstances than others. Used in digital NTSC acquisition - when the options are 4:1:1 or 4:2:2 - to produce an NTSC DVD, then 4:2:2 is vastly superior - no question. Likewise, I am certainly not denigrating it for use with any interlace system for that matter - it's why it came about. And even when used for HD (or 4K) in a progressive system (where the benefits are far less pronounced than for interlace), I'm still not "denigrating" it - rather saying that if you want better than 4:2:0, there is much to be said for 4:4:4. The worst I'd say for that is it falls between the stools of 4:2:0 and 4:4:4 IMO. You may call it a sweet spot, I'd say it falls between stools - can we just agree to differ there?

It is also not denigrating 4:2:2 to point out that in Europe, broadcasters insist on 4:2:2 (amongst other factors) for mainstream programme acquisition, uplinking etc. But 1080i/25 is the main standard for HD transmission in Europe - so I don't find that surprising, we're back to the subject of interlace. There is talk of 4K transmission in time (albeit at least a decade away) which will certainly be progressive with a framerate to be determined. Will 4:2:2 still be such a necessity then......?