Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

This might amuse Graeme...

L

laguun

Banned
Joined
Mar 5, 2007
Messages
1,732
Reaction score
0
Points
0
Yeah, makes me laugh out loud. Quite frankly, unless you've coded a high quality demosaic, which I do believe is something I can claim, you don't really understand a Bayer pattern.

There's a reason why Canon, Nikon, Sony, Leica, Panasonic, Olympus, Kodak, Hassleblad, etc. all use a Bayer Pattern Colour Filter Array in the sensors for their DSLRs / Medium format backs. That reason I outline in the "why Bayer" thread, but put simply:

Bayer pattern is the compromise (because all design is the art of compromise) that fits in best with a) real world visuals, b) human eye / brain and c) uses any given number of photosites most efficiently.

And that's the reason why no top performing DSLR uses an RGB stripe pattern sensor like the Genesis does.

No, Bayer is not perfect, but you can deal with all it's issues, it's just highly complex and hard work to do so. But that's work that I have to do, and you don't have to do. You just have to enjoy the images.

All I ask is that if anyone hires a Genesis, they point it at a 4k zone plate and see what it looks like. That will tell you the story.

Graeme
 
This is petty i know, but I noticed on a NAB video at freshdv that the panasonic (nb, not panavision) guy said the varicam used 10bit data allowing 1096 :) code values compared to the 256 of 8bit. Sounds interesting, tell me more.. . ..actually don`t..
 
And referring to sensors as 4:2:0 is just plain wrong. And quite frankly, if you think that Bayer is a 4:2:0 sensor, then RGB stripe is a 3:1:1:1 sensor, the 3 being the number of samples, and the 1 meaning you get 1/3 the resolution in red, green and blue.

So, let's do new notation:

RED is 78%:50%:50%

Genesis is 33%:33%:33%, and we'd have to reduce that number a bit for the OLPF, so I'm being generous here, and we've still not accounted for, that spatial offset in the R,G, and B, and and in Bayer, there's a maximum gap of 1 pixel before you hit one of the same colour. In RGB strip, horizontally, you have a gap of 2 pixels before you hit the same one. That's like having a much lower fill factor horizontally than vertically. At least Bayer is nicely symmetrical!

3 chip of equal total photosite count is therefore also 33%:33%:33%, and again, I'm being generous. We do have to ensure the chips are perfectly aligned though, and we can suffer from prism artifacts.

Graeme
 
It does contain some really great info on MTF and image system performance.

BTW, watch part4 2/3's in when he says: "The concept of supersampling, as one which as technology improves we will be doing more and more off. Instead of having an absolute relationship between the number of photosites and the final representation of the image, ideally we want to supersample, the same way we do in audio"
 
For me, it's just the "let's look at the issues with Bayer" without looking similarly at the issues of a prism, or the issues of a RGB stripe pattern. It's just feels a little biassed to me :-)

The whole concept of over-sampling is to push any artifacts of filters out of the visible / audible band. But that's exactly why we have a 4k sensor and get 3.2k out of it - because if we tried to get 4k out of it, we'd be in the same bad situation that any 1080p camera maker is, trying to get a full 1080 out of 1080, which inevitably leads to aliasing.

With a bayer pattern, you've GOT to be more conservative on your OLPF because we've got to minimize chroma aliasing, so we sacrifice a little luma detail too. This is a good thing, and means we don't get a 100% output. We're getting about 78% out, which means we get minimal chroma aliasing, and if you do see any, we have the tools to practically eliminate it. I've never seen any on a real world image anyway, only a test chart.

That means we can present all sub 3.2k detail without any significant aliasing artifacts and a very strong performance at the lower MTF that relate to picture sharpness perception, without the need for aggressive edge sharpening.

Some people think of a bayer demosaic as an upscaling algorithm. I often think of it as a form of compression. It's also a reconstruction filter. Notice how the subject of reconstruction filters is entirely missing from the presentation.... But that's the other half of sampling theory, and should not be ignored. I don't ignore it.....

Imaging and sensors is a complex subject. But it's also highly nuanced, and needs a good understanding of many different aspects before the picture falls into place. When I sat down at NAB and watched the demo reel, I see that we're in a pretty good place :-)

Graeme
 
Graeme, I love it when you integrate audio and video in your descriptions of sampled systems. There are so many similarities in what one does and why one does it that its pretty comforting to know that their is an underlying order to nature.

In the audio world the politically incorrect (though very accurate) term for a reconstruction filter is an anti-imaging filter. It is used to suppress the images that would otherwise occur at integrals of the sample rate.

There is a very common misconception as to the function of oversampling filters in audio digital to analog conversion. For some silly reason (likely marketing driven) many think that oversampling filters somehow add information that would otherwise, not be present (as in that which is above Nyquist). There is likely a strong analogy to what many try to do with over sharpening in the visual world, they too are trying to add back in HF detail that isn't present (or certainly shouldn't be present).

What oversampling accomplishes (in sampled audio systems) is to increase the amount of attenuation that the anti-imaging filter contributes by increasing the frequency of the images. In the end, for image suppression, what matters is the total attenuation that the anti-image filter and the oversampling filter provide when working together. Obviously, no system will ever achieve infinite image suppression, but if they are reduced in magnitude to a point where they are indistinguishable from noise, they are for all effective purposes gone.

I think that the same can be said of optical systems, if the performance of the encode and decode system is such that the aliases are indistinguishable from noise, well then its about the best you can hope for.

As for the demo at NAB, I too agree, you are in a very good place indeed!
 
Ah, now its getting funny - an coworker just told me that Panavisions very own Andy Romanoff has posted their spin & bad science in cinematography.com, guesss where - not in the Genesis, but in the RED subforum. hrrhrr.

http://www.cinematography.com/forum2004/index.php?showtopic=29708&st=40&start=40

I am not to sure what panavision intends to achieve by posting wrong information which skips completely kodaks (and hasselblads, canons, arris, reds, nikons...) work and science of the recent decades.
 
Laguun the thread is titled "Sony wanted the Genesis to be 4K; , John Galt insisted on 1920 x 1080" I can see why someone from Panavision would want to post the presentation there.
And the presentation although it does favor the Stripped system this is because Panavision like the stripe system just like Graeme likes they Bayer system I doubt either would say "Well this other system is better but we chose the latter because we're stupid".
 
Look at the striped system in context of what they're trying to do: they're using 1920x3 to deliver 1920. They are getting 4:4:4 color sampling for a 1920x1080 image. They're not trying to do 4K, that's the part that's being left out. There's nothing about "spin" or "bad science" there. It's just the path they chose.
 
Barry Green said:
Look at the striped system in context of what they're trying to do: they're using 1920x3 to deliver 1920. They are getting 4:4:4 color sampling for a 1920x1080 image. They're not trying to do 4K, that's the part that's being left out. There's nothing about "spin" or "bad science" there. It's just the path they chose.
Click to expand...

It's the same scenario in the 3 chip system as well.
 
Barry Green said:
Look at the striped system in context of what they're trying to do: they're using 1920x3 to deliver 1920. They are getting 4:4:4 color sampling for a 1920x1080 image. They're not trying to do 4K, that's the part that's being left out. There's nothing about "spin" or "bad science" there. It's just the path they chose.
Click to expand...

So it's kind of like a three chip camera with pixel offset... because it has 3 sets of 1920 x 2160.... it's just they're not in line with each other... right? The only thing I'm not getting is the 2160... I know it's twice 1080... but why? To me it would make sense to be a 5760 x 1080. Or are they grouping 2 vertical cells together?

CraigWB said:
It's the same scenario in the 3 chip system as well.
Click to expand...

Right, but with it being a single sensor in that set-up they can use 35mm lenses with 35mm dof.


For a 1920 x 1080 camera with 35mm dof... I don't think they could have choosen a better system. The only thing is you're not getting anything above an HD resolution.
 
Andrew McCarrick said:
So it's kind of like a three chip camera with pixel offset... because it has 3 sets of 1920 x 2160.... it's just they're not in line with each other... right? The only thing I'm not getting is the 2160... I know it's twice 1080... but why? To me it would make sense to be a 5760 x 1080. Or are they grouping 2 vertical cells together?
Click to expand...

I think its in line with their whole super sampling idea, they talked a little bit about how other cameras super sampled horizontally but not vertically I don't recall the exact argument but I think it had something to do with that.

And I do think that it would be possible to do a 3 chip camera at 35mm, my understanding is it's more of a cost issue than anything 3 strip technicolor worked after all.
 
Barry Green said:
Look at the striped system in context of what they're trying to do: they're using 1920x3 to deliver 1920. They are getting 4:4:4 color sampling for a 1920x1080 image. They're not trying to do 4K, that's the part that's being left out. There's nothing about "spin" or "bad science" there. It's just the path they chose.
Click to expand...

Sorry, with all respect, i disagree.

Describing a bayer pattern CMOS sensor without not explaining what Dr. Bayer and decades of science since then brought to demosaicing, and why the huge majority of all top-end sensors and all top-end camera manufacturers today use this method, is spin, or naive, or both.

Judging the quality of digital interpolators takes a little more than depicting a pattern - thats where the maths -starts-, and thats where his presentation ends.

I like the Sony cameras which Panavision rents out, in fact i bought a hdcam pretty early. But intentionally (or due to not understanding, which i dont think) misinforming the audience is bad marketing. Without going to much into detail - there is a reason, why Hasselblad, Kodak, Arri, Nikon, Red, Dalsa, Canon etc use CMOS. And btw, different from Panavision, they design their digital cameras themself.

The presentation part 1-5 and 7 is good, only the part where hes trying to falsely depict -most- competitors using CMOS isnt complete, and thats not all to professional. If you talk about competitors, then do it correct.

There is a -clear- reason why they only show theoretic figures instead of actual pictures of charts - and thats because Dalsa and Red easily resolve more and w/o aliasing than any uncompressed 1080p camera, including our sonys here.

The Sonys and Panavision 1080P cameras (750/790/900/23/Genesis/35) have advantages over the red - resolution clearly isnt one of them, in that aspect they are inferior.

He also has more factual mistakes in the presentation (regarding sony, arris and thomsons sensors), but i wont go into detail there. Its easy enough to check the correct numbers on the websites of the respective manufacturer. He also misses the important fact, that with raw cameras, the artists and/or DIT can choose its approach of more or less resolution, more or less aliasing etc himself - not being limited by fixed systems as in panavision and sony. Arri, Red, Thomson, Dalsa choose the other way, which has certainly not only advantages, but offer more control over the process than sonys approach, as found in genesis.

Anyhow, the markets already have decided, red outsold the competition by far, and if panavision intends to stay with 1080p its fine, we are using 1080p all the time. There are however *many* scenarios in which a higher resolving camera and recodring is pretty helpful, even when mastering for 1080p or 2K.

Its always better to simply outsell, outperform and underprice the competition than to make look them bad - that can backfire, and i will certainly use panavisions presentation by stopping after part 6, then continuing by "what they -forget- to tell...." and showing actual shots and charts.
 
CraigWB said:
I think its in line with their whole super sampling idea, they talked a little bit about how other cameras super sampled horizontally but not vertically I don't recall the exact argument but I think it had something to do with that.
Click to expand...

well now that I look at the video again it looks like they're grouping 6 pixels together. Two of each color in a 3 x 2 pattern and basically rendering that as one pixel (with 3 color channels; so 1920 x 1080).... but achieving the 35mm dof.

So Red, Dalsa, ect. are achieving more resolution with the same space via the Bayer Pattern... but Panavision is getting the necessary resolution they wanted but making it so you can use the 35mm Panavision glass that already exists (without the need for new lenses with a prism.).
 
Yes, I fully understand why they chose the system they do, and it's a good compromise for what they want to achieve. It may be "4:4:4" for 1080p, but it's only 33%:33%:33% (generous rating) compared to the sensor. We get a full 4:4:4 from our 2k from 4k, and we have the benefit that you can choose your own downsample filter, which I think is important. With the stripe patterns and on-chip binning, there's very little choice in what you do with the data.

Graeme
 
To say that RED is more "successful" than the Panavision Genesis camera because it sells more units is not particularly accurate since the two companies don't even follow the same business model. You can't buy a Genesis even if you wanted to. And even if you compare a RED to a camera that you can buy, like an F23, it doesn't necessarily mean that a company like Sony won't make some money off of their product even if the sales figures are lower. It's like comparing the sales of a Toyota Camry to a Ferrari -- of course more Toyotas are being sold. That's not necessarily a problem for Ferrari IF they don't lose their customer base (not to say that Sony shouldn't be worried, just that it's not simply about number of units sold.)

It's like this strange argument that digital is more successful than film because more digital cameras and projectors are being sold than film ones. That's been true for years, so what?

That completely ignores the fact that a lot of movie production follows a model of rental, not ownership, and the fact that movie theaters routinely use decades-old 35mm projection equipment. I can go out and shoot a feature right now on a thirty-year old Panaflex or Arriflex and as long as I put a modern lens on it, it would look no different than a movie shot on a one-year-old Panaflex or Arriflex. So the fact that more digital equipment is sold doesn't necessarily follow that more productions will soon be shot digitally, because you can rent older film equipment for years to come -- and also, digital equipment has a faster obsolescence rate so of course the rate of sales will also be higher and faster.

And no one seems to notice that Panavision is doing the digital world a big favor by making digitally-shot movies so available for mainstream audiences to see and accept, get used to, and make producers and studios comfortable with digital acquisition and distribution. Right now we have in theaters, or recent to home video, several movies shot on the Genesis: "The Other Boleyn Girl", "The Forbidden Kingdom", "Deception", "21", "Walk Hard"... so more people are going to see movies shot on the Genesis in 2007 and probably 2008 than people will see movies shot on the RED, despite how many RED cameras are sold (and not all are being sold to feature filmmakers.)

The whole competition between RED and Panavision is a bit silly if you ask me. Panavision has a rental-only business model and is heavily tied to the Hollywood studio system, because they provide consistent and good service to producers that can afford them. RED is a camera for sale and while no doubt some Hollywood production will (and has already) use the RED, it will have even more impact in other areas or countries where camera ownership is a more common model for production. Sony and ARRI probably have more to worry about RED than Panavision.

As for John Galt, he's a bright man who knows more than I do, but to some extent, he's also defending a product he helped create (why shouldn't he?), and he has some salient points to make about how pixel resolution is not necessarily the way to measure true image sharpness. His comments can be more damning about film resolution in some ways than they are against other digital cameras.

Rather than get into the numbers game, I'd suggest people look at real world results and judge for themselves.
 
You must log in or register to reply here.
Back
Top