Welcome to our community

Be a part of something great, join today!

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 :)

Panavision videos...

so here's the answer jim you have to buy panavision take their lens secrets and start selling us the glass kill off the genesis and replace it with epic :)
 
pvillow said:
Check it out here:
http://media.panavision.com/ScreeningRoom/Screening_Room/Demystifying_Part6.html
Click to expand...

I was so inspired by Part6 I made my own chart.

582_1209657085.jpg

If it walks like a duck and quacks like a duck, it's fuckin duck.

Seriously, Panavision needs to fire this guy and hire someone who did some research and found out how a Bayer pattern actually works.

Ian
 
I don't think I watched the same videos you guys did. What I saw was no not as bad as you guys are making it out to be.

There were some flaws and some points they didn't expound on and they omitted some things, but they didn't attack anything. There are not many people who can stand in front of a room full of people and talk about a subject without putting their interpretation on it.

Saying Bayer is only 4:2:0 was the big mistake I thought they made and they said they were going to talk about it more and then never did. I could have understood an argument for 4:2:2, but never 4:2:0. And I know 4:2:2 refers to something completely unrelated to sensors.

They also explained in their own presentation how there is data past Nyquist, but that data has problems. They also talked about how design of the OLPF in a system is important to maximize resolution and limit aliasing and that the OLPF could be designed so that it gives the maximum green channel and limited aliasing in red and blue.

I think you guys need to relax a little. If you don't like there presentation, make you own.


Mahalo,
Dusty
 
Yes, a lot of the science stuff was very good and very well explained. My big concern (other than what I've written above) is that 4:2:2 or 4:4:4 or 4:2:0 is not an appropriate terminology for a sensor and is highly misleading.

And that OLPF optimization is an issue, but we address it to our satisfaction, in both hardware (OLPF) and software (chroma de-moiré tools), but quite frankly, most 3 chip cameras allow for more aliasing (even when they use an OLPF) than I'd feel comfortable in a high quality imaging product.

Graeme
 
Plasmatic said:
Meanwhile I've been told they haven't shown an operating profit in over 20 years. I bet they would if you were running the show. They'd probably have to do without a few dozen less vice-presidents or so:-)
Click to expand...
Welcome to industry bookkeeping. None of the major film studios show an operating profit, either. No profit = tax write-off.
 
My big concern (other than what I've written above) is that 4:2:2 or 4:4:4 or 4:2:0 is not an appropriate terminology for a sensor and is highly misleading.
Click to expand...
Is it that misleading though? I mean, I know in general it is, obviously -- they're trying to paint the scenario that a 1920x1080 image is better than a 4096 image, so in that perspective of course it's misleading. And a sensor in and of itself doesn't do color sampling, so it's wildly inappropriate, but -- at its core, is the overall simplified message being that misleading?

In context with what the sensor is designed to do, were they intentionally misleading? Their sensor is designed to deliver a 4:4:4 @ 1920x1080 image. They show how they accomplish it. The Bayer sensor cannot deliver a 4:4:4 @ 4K image. It can't. It can deliver a 4:4:4 @ 2K image, yes, but not at 4K. So their position seems to be (consistent with what they've said before) "you can't get 4K out of a 4K sensor". And you've agreed with that. So while it's alarming (and inaccurate) to label the sensor itself as 4:2:0, is the ultimate underlying message inaccurate? The underlying message being: "you can't get a full 4K from a 4096-wide Bayer-pattern sensor, and you certainly can't get 4K @ 4:4:4 from a 4096-wide Bayer sensor." Which I think we can all agree with.

Here's the part that galls me though. You can't get 4K -anything- out of a Genesis. So, we have a system that's designed to give you a solid 2K of detail at 1920x1080 @ 4:4:4. Fine, the Red can do that too. But the Red can also give you a solid 3K of detail, at something probably close to 4:2:2. The Genesis can't. And the Red can even give you a 4K frame size with 3.2K of resolved detail, at probably somewhere between 4:2:0 and 4:2:2. The Genesis can't even begin to think about that. So your Bayer sensor does everything their striped sensor does, yet it also does things their striped sensor cannot, and they're somehow trying to portray that as a problem? That's the misleading/deceptive part, AFAIC.

Where they really go astray is using reasonable arguments to lead the audience to an unreasonable conclusion. If all you're looking for is 1920x1080, then the Genesis approach looks pretty good -- but the Red approach looks just as good. However, if you want more than 1920x1080, the Red offers that and the Genesis doesn't, and to totally ignore that is irresponsible on their part. To imply that the Bayer approach is "bad" because it offers MORE than the Genesis? That's just kooky.
 
For all the mathematical calculations going on here isn't the most accurate number to use 3.2:2:2? I know Graeme doesn't think this terminology should be applied to a bayer sensor (And he's right) but that won't stop the world from doing so. So let's just say that the Mysterium is 3.2:2:2 Which we all recognize as 3.2K Luma res, 2K chroma res and be done with it. It still kicks the pants off 4:4:4 1080.
 
Evin Grant said:
For all the mathematical calculations going on here isn't the most accurate number to use 3.2:2:2? I know Graeme doesn't think this terminology should be applied to a bayer sensor (And he's right) but that won't stop the world from doing so. So let's just say that the Mysterium is 3.2:2:2 Which we all recognize as 3.2K Luma res, 2K chroma res and be done with it. It still kicks the pants off 4:4:4 1080.
Click to expand...
Yeah, but the thing is it's a numbers game.

Whatever the sensor "actually" is, it's certainly "effectively" 4:4:4, as anyone who has pulled a competently-lit greenscreen key from RED footage knows. I have never seen anything -- including film -- have as close to a one-click key as RED footage does.

To call it any less than 4:4:4 -- while perhaps mathematically accurate -- ultimately sells it far short of what the final imagery accomplishes. People don't really know what those numbers mean, they just want to see 4s across the board. Anything less is, well, less.
 
But Barry, you can't get a full 1920x1080 from a 3 chip camera who's chips are 1920x1080 unless you allow for significant aliasing as optical low pass filters just are not steep enough. Say you can get a good 90% even, then you're a 3.6:3.6:3.6 image now ;-), and even that is an abuse of chroma sub-sampling notation as chroma sub-sampling notation DOES NOT tell you anything about the actual measured resolution of the luma or chroma, just by what ratio the chroma is sub-sampled compared to what the luma is sampled at.

And, given the segment on Bayer demosaicing, which seemed quite a simplistic look at a complex subject, there was no mention on how the lack of co-sitedness of the RGB in the stripe pattern is dealt with, and also no mention of the effective fill factor in a RGB stripe being greater in the horizontal direction than the vertical direction, and how that effects OLPF design.

So that's why I saying that at 4k we're 78%:50%:50% (as percentage measured resolution of the of the system) and that "at best" a Genesis is 33%:33%:33% of it's "~6k" sensor. I'm sorry, but 50% of 4k is better than 33% of < 6k (by a small amount) and 78% of 4k is vastly better than 33% of <6k.

What we need next is a good metric of aliasing based upon both charts and real world images.

Graeme
 
Scott, the image is reconstructed to a 4:4:4 RGB image, and hence that's how it keys. We take great care to ensure our edges are "nice".

Graeme
 
I think the presentations were very good and informative. Red should organize similar presentations of its own to educate the people. There is a lot of disinformation and misinformation on this forum. Red's representatives do try to dispel some of it. However, more is needed in terms of explaining their digital vision in a regulated and streamlined fashion, much like the way Panavision people did.
 
I found the whole presentation very interesting. It was nice to learn more about MTF, and how film isn't all it's cracked up to be once you get all the way to projection.

Once he started applying x:x:x terminology to single Bayer sensors, it really broke down. First he said it was impossible to measure the resolution of a single Bayer sensor, then started applying the x:x:x terminology and talking about how diagonally it was 4k. Lots of poor assumptions, if you ask me.

Yes, I would think it is impossible to measure the resolution of a single Bayer sensor... using the same technique as a 3CCD system. It doesn't hold up, but there are other ways.

I'd like to ask specifically about OLPF in regard to Bayer.

It seems to me you do not have to create a special OLPF that accomodates the RBG Bayer pattern, as though you have to treat the green photosites differently than the red and blue ones (as was inferred in the presentation). Really, wouldn't you treat the sensor as monochromatic, since that is what you really get out of the sensor? You'd just create the right OLPF for a monochromatic 4k sensor, based on size of the photosites and fill factor, right?

The goal is still to limit aliasing while maximizing remaining resolution, but you can only fairly compare a single Bayer sensor to a single CCD sensor.

The fact that the Genesis photosites are not co-sited, but next to each other in stripes was disconcerting to me. How do they deal with that? It seems like they need their own "debayer" like process to reconstruct an image, considering they are not co-sited. I can only imagine what would happen if they simply treat those RGB's as though they were co-sited. Again, you'd get nasty aliasing/artifacts, right?
 
Absolutely Joofa - but question - what do I with the rest of my time ? :-)

Yup, if you treat non-co-sited colour filtered pixels as co-sited, you do get aliasing / chroma fringing. Unless you OLPF strongly enough.... Or do some kind of process in software.

Graeme
 
Graeme Nattress said:
Absolutely Joofa - but question - what do I with the rest of my time ? :-)
Click to expand...

Yes, I know you, and the rest of Red team are busy now. However, in my estimation Red should take a more proactive stance than a reactive one by alluding to the use of Steven Soderbergh / Peter Jackson as a testament of Red's ability, and display at trade shows. I know very well that reference to Soderbergh and Jackson does have a resonance with certain segment of the film community. However, Red should be able to lay down the facts via presentations, and demonstrations that go beyond NAB, so that independent-minded people can make a decision. Additionally Red's website should be more streamlined. It must have a section on Human Resources and Careers even if does not intend to hire any new personnel at this stage. That gives an impression of a company serious in growth as opposed to just an online website for commerce/ordering.

Red must promote itself as a company where serious R&D is held in addition to being an online store as the current website suggests.
 
I think you guys need to put out more Myth Buster episodes too. FXPHD is doing a couple REALLY good RED classes, but you have to pay to get in on those, so it isn't for everyone. It would also be better for more education to come from the company behind the camera too I think.

You CAN get all the info you need by sifting through REDUser, but that's just it, you have to sift through a LOT of stuff, and there's a lot of garbage too. I really liked the first Myth Busters episode that you have out, and think that it would be easy to point someone who is skeptical about editing RED footage to that video to prove it's really really easy. You should really make more of those episodes and make sure they're easy to find.
 
Joofa said:
Red must promote itself as a company where serious R&D is held in addition to being an online store as the current website suggests.
Click to expand...

Red already is promoting itself as a serious R&D company. Just to name a few "promotions": Red One, Scarlet, Epic, RedRay... :)
 
Jason Ing said:
Red already is promoting itself as a serious R&D company. Just to name a few "promotions": Red One, Scarlet, Epic, RedRay... :)
Click to expand...

Jason, an end user does not have a visibility to where/who exactly is making these products. We do know that Red has some inherent R&D ability (Graeme Nattress/Deanan/Stuart English are there), but what exactly is indigenous is not known at least to me. I do not expect in todays world for a company to make/install all its parts (does Dell make that processor in its computers?). However, one has some idea about other companies that what they are up to.

Red's website is just a front end and things stop there. I was making a point about that if Red's R&D is something they are interested in *growing* (and is not all OEM) then they should provide some visibility to prospective people out there and means for that people to apply to be a part of that group.

You go to any well-established company's website and you see a career section. Red should not miss out on that.
 
I think one of Mr J's more proven points in Oakly, and now RED, is his ability to hunt down the right people for the organization... Maybe he likes it that way?
 
You must log in or register to reply here.
Back
Top