Creative Cow Article: Panavision, FUD and...HDR?

[Cal Brunker]

So, Creative Cow just published an article with John Galt of Panavision:

"The Truth About 2K, 4K and The Future of Pixels" Link

Mostly, it's more of the same FUD as his previous "Bayer sensors are baaaaad" dog & pony shows. Yes, everyone knows that you don't truly get 4k out of a 4k Bayer sensor. Everyone knows that an OLPF is a necessary compromise when using a Bayer sensor. He, of course, fails to mention the issues with non-colocated photosites in the Genesis camera. Anyone who's been following Red's development for the last several years and has half a clue knows all this already, and it's mostly a non-issue.

There were, however, 2 things of great interest in the article. First, John make a very salient point about resolution (shocking, I know): almost all normal theatrical installations are going to be 2k for the time being, and (considering how crazy theater owners are about the costs of moving to digital in the first place) that means for the foreseeable future. Shooting 4k and up still has advantages in post like shot reframing, and additional resolution for all manner of FX work. But, the final delivery medium will typically 2k or less, exclusive of IMAX or whatever the electronic equivalent of IMAX turns out to be.

My take is that's nothing but good news for Red. 4k is already plenty of resolution for anything up to theatrical presentation in the near term. Anything more is just gravy for special applications or large-venue. The point that higher framerates may be the future is valid, as well. I think James Cameron's right on this one: there's more perceived visual bang for the buck by bumping framerates to 48fps than any other single change in motion picture imaging.

Second, and perhaps even more interesting, is the announcement of Panavision developing an HDR sensor they're calling "Dynamax". From his description, it seems there are 6 photosites for each pixel. So, is that 6 non-colocated photosites for each pixel in...what? A striped, or (gasp) Bayer pattern? Or, is it some kind of Foveon-like layered configuration?

Oh, now I see - it's some kind of diagonal-pixel, row summed technique called "3-pixel binning". Wouldn't this introduce the same kind of artifacting as Bayer sensors? What kind of OLPF would be needed for this monster?

Panavision Link

Dynamax is new to me - I apologize if I'm behind the curve. I didn't find a single hit on a search for "Dynamax" here on Reduser.

I have noticed that Reduser has basically died 'cause the team at RED has given us the baddest ass camera on the market and we're out shooting stuff. We've cut our time at this site by drastic amounts since our cameras arrived.

It's nice to see that the people making harsh criticisms of the RED one have lots of time to do so, 'cause they have nothing they can afford to shoot on.

P.s. I'd like to see the look on Steven Soderbergh's face if you came to him whining about Bayer pattern sensor resolution. The subtitle under his expression might say "who are you, stop bothering me, I'm making a movie".

Yeehaw.

Cal

 

[Jonathan Stevenson]

For some reason, posting the link into the forum doesn't work... Here's the link to the Google search, it's the first result.


http://www.google.com/search?client...eative+cow+future+of+pixels&ie=UTF-8&oe=UTF-8

 

[chuck colburn]

For some reason, posting the link into the forum doesn't work... Here's the link to the Google search, it's the first result.


http://www.google.com/search?client...eative+cow+future+of+pixels&ie=UTF-8&oe=UTF-8

Thank you.

 

[Shawn Nelson]

Wow, is that true? For being an engineer, I've largely stayed out of a the 'Bayer pattern' debates. Is the Red's "4k" really less actual information than the Genesis?

 

[Rick Darge]

I don't see the point. I mean, I understand what he is saying but I keep thinking, 'Who cares? My Red makes beautiful images. I don't give a crap if its TRUE 4k or not. It takes amazing imagery and that's all that matters'

 

[Jonathan Stevenson]

I don't see the point. I mean, I understand what he is saying but I keep thinking, 'Who cares? My Red makes beautiful images. I don't give a crap if its TRUE 4k or not. It takes amazing imagery and that's all that matters'

I agree. It doesn't really matter as long as the images are slick, which they are!

 

[mikeburton]

Interesting to hear Panavision is working on a HDR cam. Not surprised though, people have been asking for this kind of camera for years now since Paul Debevec introduced HDRI photography for 3D lighting in 96-97.
It will be interesting to see how things develop on that front.
What's even more interesting will be cameras in the future that can extract a z-depth channel, and multiple alpha channels ready for compositing where everything can be instantly manipulated in post. This sounds rediculous right now but it will happen.

 

[Steve Sherrick]

Wow, is that true? For being an engineer, I've largely stayed out of a the 'Bayer pattern' debates. Is the Red's "4k" really less actual information than the Genesis?

Different approaches to imaging. Each one with pros and cons. Each side with an agenda. What's important is how it looks to you. I personally believe each side has a valid argument, but one fact remains - Genesis is a camera you rent, Red is a camera you rent or buy. That doesn't make Red better, just places it in a different space. Panavision has done some very remarkable things over the years and they continue to work closely with cinematographers to see what they want. It's not a bad model in my opinion. Red has appealed to a slightly different group, with some crossover to the Panavision crowd as well. I think they can co-exist, have different philosophies, and in the end the filmmakers win.

Hopefully this doesn't turn into a mean-spirited Red vs. Panavision thread. John's article is informative, opinionated, and offers healthy debate. It would be great if that's how it was treated here and the debate remains positive, perhaps with contributions from the Red team to offer their take on things.

I personally find the science behind all of this intriguing and want to learn more from guys like John and Graeme.

 

[Alex.Mitchell]

Is his name seriously John Galt? Imagine being named after an Ayn Rand character. Cripes.

Anyway, I think that there are setbacks to every approach in imaging. I've poured over the white papers and, honestly, it's a fun topic to debate academically but there's not much of a point considering how good the images look.

 

[Graeme Nattress]

It's been discussed to death before... The main issues, to me are that:

Theoretically, there are problems and compromises with the Bayer colour filter array approach. However, practically, they're less of an issue, and very practically don't actually occur during recording of real world imagery. Bayer CFAs can be incredibly complex to deal with, but you don't have to deal with them. That's my problem. They are, however, the most efficient way to arrange a finite number of photosites in a perceptually relevant manner.

RGB Stripe pattern sensors are much less efficient in arranging photosites in a perceptually relevant manner. The differential spacing horizontally theoretically causes all manner of imaging problems with regards to aliasing. John rightly says " But you have to use an optical low pass filter to make sure that you don't create a moire pattern." But that also applies to any sampled system or camera - 3 chip CCD cameras, 1 chip RGB Stripe Colour Filter Array cameras you name it. All of them need them.

In a Bayer CFA camera we know that the three colours of photosites are not co-sited, and we act accordinging. Indeed, we use the spatial offsets of the red, green and blue photosites to extract real resolution from them. It's not magic, it's just making the most of the data you've got.

In a RGB stripe CFA camera, we also know that the three colours of photosites are not co-sited, but that is often ignored. The spatial separation is not used to extract real resolution from the data.

The best answer though, is why if the Genesis is a 6k camera, does not the resolution measure somewhere near 6k.

As I said, if John applied similar scrutiny and scepticism to the RGB Stripe Colour Filter Arrays as he does Bayer Pattern CFAs, it might make for an interesting discussion...

Graeme

 

[Tonaci Tran]

It's definitely a marketing piece to keep them going..to battle the "hype."
The HDR thing they are working on sounds really interesting though.

 

[Graeme Nattress]

Sure it's interesting, and without the hype and negative marketing it would make for an interesting news story or white paper. But why bury such interesting news deep in negative article?

Graeme

 

[josemnorton]

I don't see the point. I mean, I understand what he is saying but I keep thinking, 'Who cares? My Red makes beautiful images. I don't give a crap if its TRUE 4k or not. It takes amazing imagery and that's all that matters'

I actually care but I'm not gonna loose sleep as long as I keep getting the amazing pictures the RED gives me!

Thanks for all your replies.

 

[Matt Gottshalk]

Sure it's interesting, and without the hype and negative marketing it would make for an interesting news story or white paper. But why bury such interesting news deep in negative article?

Graeme

That was my thought too...seemed like one big negative ad.

 

[Joe G.]

I also noticed his little digs spattered throughout the article.

His attacks were mostly aimed at supposed claims, or hype, and not the actual performance of the CMOS sensors. Red admits that the effective resolution is somewhat less than 4k. I believe 3.2k was tossed around. This is still very useful, very sharp.

The HDR option is where he seems to be focused now. Is Red going to look at something similar? Fuji cameras did something like this, and they now have a sensor with two different pixels, or photosites paired off across the sensor. One big, one little. It's supposed to increase dynamic range, but I haven't seen it tested and compared.

Obviously this is the next frontier. Is Red going to respond?

Or is the Monstro going to have enough dynamic range to fend off the challengers for a while?

When are we going to see some Monstro action?

 

[conrad gaunt]

It's been discussed to death before... The main issues, to me are that:

Theoretically, there are problems and compromises with the Bayer colour filter array approach. However, practically, they're less of an issue, and very practically don't actually occur during recording of real world imagery. Bayer CFAs can be incredibly complex to deal with, but you don't have to deal with them. That's my problem. They are, however, the most efficient way to arrange a finite number of photosites in a perceptually relevant manner.

RGB Stripe pattern sensors are much less efficient in arranging photosites in a perceptually relevant manner. The differential spacing horizontally theoretically causes all manner of imaging problems with regards to aliasing. John rightly says " But you have to use an optical low pass filter to make sure that you don't create a moire pattern." But that also applies to any sampled system or camera - 3 chip CCD cameras, 1 chip RGB Stripe Colour Filter Array cameras you name it. All of them need them.

In a Bayer CFA camera we know that the three colours of photosites are not co-sited, and we act accordinging. Indeed, we use the spatial offsets of the red, green and blue photosites to extract real resolution from them. It's not magic, it's just making the most of the data you've got.

In a RGB stripe CFA camera, we also know that the three colours of photosites are not co-sited, but that is often ignored. The spatial separation is not used to extract real resolution from the data.

The best answer though, is why if the Genesis is a 6k camera, does not the resolution measure somewhere near 6k.

As I said, if John applied similar scrutiny and scepticism to the RGB Stripe Colour Filter Arrays as he does Bayer Pattern CFAs, it might make for an interesting discussion...

Graeme

I agree, but to be fair, the demosaic process reduces signal to noise ratio, when compared to a theoretical 4:4:4 equivalent image, which isn't quite the same as "extracting real resolution from the data", as you hint is possible.

A good debayer algorithm should be statisically correct with its educated guess more than 50% of the time, when we calculate the values of the missing data, by analysing the surrounding discrete samples. Sure its better than striping.

I love the red, but I always downsample to 2k from TIFF when i need quality, robust channels, especially the blue channel, and this also helps with the compression issues, which are subtly there.

1:1 zoom into the debayered image aren't nearly as hot, again MO. Thats not a complaint, just what I do if I need to apply heavy grading.

1:1 resolution is fine for real-world, natural history, documentary stuff, obviously.

Only after a good downsample IMO, does the extra spatial resolution (luminance) put it above HDCAM in terms of eye-candy (technical term). A 1080 crop from the center, or 2k, doesn't compare. I have clients who agree.

I strongly disagree with the "leave the debayering to us" approach also, and find it a little patronising and disappointing.

Access to raw data is highly desireable for many reasons, not least better archiving, faster motion tracking, super-resolution algorithms can gleem more from the original bayer pattern, and not least speeding up/improving transcodes to Cineform RAW (should you decide your customers know what they want)

I've read that its jpeg2000 wrapped in meta-data. That people are decoding it using OpenJPEG on linux. Commercial APIs are used so its top-secret...

I don't know, and I care less. Jim built the Red after getting annoyed at having to use a proprietory tool to get at his video data right? If thats true, its a little ironic..

Well I don't care what flavour of wavelet file is used (and there's plenty btw, some lossless AND fast), I'd just like the raw data, transcoded into an open format that can be cropped and trimmed, cos migrating "dead weight data" is "S**t".

6k is better than 4k, period, whatever method is used, fact. 28k will be better still no doubt..

And the sensor layout used, a secondary issue in comparision..

 

[David Rasberry]

I think Galt's observations about actual display resolutions at normal viewing distances are pertinent. 2k digital projection when viewed from appropriate distances, more than 2 times screen height or so, is as good as most of us can visually resolve with our eyes. Dynamic range and color contrast are more important.
I also thought Panavision's corporate philosophy of manufacture for rental vs purchase was interesting. They are not about to let anyone steal their cheese.
But the film/TV industry is far from monolithic. All of these camera systems and business models exist because they fulfill a purpose that has a market. When the market shifts or ceases to exist, the businesses and products will change or disappear.

 

[tj williams]

Will the Epic Sensor Beat Panavision?

Will the Epic Sensor Beat Panavision?

Here is a very interesting article I hope is being red by the RED Tech gurus!!!! Written by John Galt who did the Panadigital stuff. I eliminated the link Just go to the current creative cow magazine
TJ

"DynaMax is designed to do that for moving images. With those 6 red, 6 green and 6 blue photosites for each output pixel, you'll have the equivalent of shooting 6 images with different exposures at once, and blend them together to create a single high dynamic range image. You'll be able to capture extreme highlights, the near highlights, the mid highlights"....

"The DYNAMAX-35 sensor is a multimode video sensor capable of operating up to 120 fps at 6x HDTV and 30fps at full res of 37Mpix"

 

[Tim Whitcomb]

dead link FYI

 

[michael zaletel]

Link is:

magazine dot creative cow dot net

/article/the-truth-about-2k-4k-the-future-of-pixels


-shooter

Forum software seems to have a bug with the real link as I've tried to paste it a few times and keeps coming up like first post dead link.