Jump to content

Recommended Posts

  • Premium Member

The problem is that they're not calling it a 9.5MP camera, they're calling it a 4K camera. This terminology has traditionally referred to systems such a film scanners which used cosited RGB samples. It is impossible for a 4000-pixel bayer array to derive the same amount of information as a 4000-pixel cosited RGB array, even without compression, (and there is a lot of that) and in my view this makes the claim dishonest.

 

What Graeme then did was to try and justify the 4K claim by pointing out that the red sensor is more than 4000 photosites wide. This invalidates his argument about compression ratios, even if I did need to reevaluate my position based on an error of one twenty-fourth in the maths - which I don't. Unfortunately, you can't have it both ways.

 

The crying shame of this is that what red have done would still be impressive if they'd just been upfront about these things, and all the other things they've claimed then not done, or done late, or tried to obfuscate. If they'd been straightforward about it, it'd have been easy to like a reasonably good 2K camera for that price. I have never claimed that red was 2K and was therefore a bad camera. Now, to be fair, I think that until the recent sensor refresh it was a fairly bad camera, noisy and clippy all at once when really one is normally expected to offset the other, but this is in no way unexpected given what they tried to do with the technology available. I could have told them that in 2006 (and I did), not that they should have needed me to. They did it anyway. They were warned.

 

And I don't prticularly rate F35 or Genesis, either.

Link to comment
Share on other sites

The problem is that they're not calling it a 9.5MP camera, they're calling it a 4K camera. This terminology has traditionally referred to systems such a film scanners which used cosited RGB samples. It is impossible for a 4000-pixel bayer array to derive the same amount of information as a 4000-pixel cosited RGB array, even without compression, (and there is a lot of that) and in my view this makes the claim dishonest.

 

What Graeme then did was to try and justify the 4K claim by pointing out that the red sensor is more than 4000 photosites wide. This invalidates his argument about compression ratios, even if I did need to reevaluate my position based on an error of one twenty-fourth in the maths - which I don't. Unfortunately, you can't have it both ways.

 

The crying shame of this is that what red have done would still be impressive if they'd just been upfront about these things, and all the other things they've claimed then not done, or done late, or tried to obfuscate. If they'd been straightforward about it, it'd have been easy to like a reasonably good 2K camera for that price. I have never claimed that red was 2K and was therefore a bad camera. Now, to be fair, I think that until the recent sensor refresh it was a fairly bad camera, noisy and clippy all at once when really one is normally expected to offset the other, but this is in no way unexpected given what they tried to do with the technology available. I could have told them that in 2006 (and I did), not that they should have needed me to. They did it anyway. They were warned.

 

And I don't prticularly rate F35 or Genesis, either.

 

We clearly stated what the the measured resolution would be of the camera system right from the start, and indeed, were actually a bit conservative on that figure as in practise we have achieved more. That is not dishonesty. We have always described the nature of the sensor and colour filter array pattern used, number of pixels, measured resolution of the camera system as a whole.

 

Just as when you do a 4k scan on film, you don't actually see 4k measured resolution from 35mm film - it's somewhat less, and different in each of the RGB channels too, the 4k refers to the container rather than the actual measured resolution of the image content. Similar with video where even VHS had the same 576 or whatever lines as the broadcast TV used for picture information, but the horizontal measured resolution was somewhat less, or HDCAM where one of the compression steps was to reduce the 1920 to 1440, or DVCProHD where they did 1280 to 960. Even with cameras, the only way that can achieve a measured resolution equal in size to the container format is to allow for such excessive aliasing as to render such resolution advantage useless. So, the measured resolution in percentage terms compared to the container resolution for RED is similar (often better) than that of 1080p cameras.

 

"And I don't prticularly rate F35 or Genesis, either." Ah, finally something we can agree on! I knew it would occur one day, but wasn't quite prepared for it to happen today. We probably agree on other things too...

 

Graeme

Link to comment
Share on other sites

Okay, now I'm trying to understand the difference between Bayer and co-site.

 

After reading around, it seems both have an array of photosensors overlaid with a grid of color filters.

In the case of co-site sampling, to make a single image several exposures are combined with the sensor being moved between exposures, so that every point is scanned at least once in each of the three colors.

 

With the Bayer system a single exposure is made so that each point is only scanned once. Colors are interpolated from neighbouring photosensors?

The Bayer pattern repeats on a 2x2 grid, so a naive first guess would be that however many photosensors there are on a Bayer chip you'd have to divide by 4 to get the number of color pixels it produces. I'm sure the number 4 there is wrong, but I don't know what the correct number should be. I'm pretty certain it would have to be strictly greater than 1 and no more than 4 but that doesn't narrow it down much!

 

To put some numbers on that, say there was a 1000x1000 photosensor co-site system. Then the number of full color RGB pixels you'd get out would be 1,000,000.

 

If you had a 1000x1000 Bayer system, then the number of full color pixels you'd get would be 250,000?

Or maybe 333,333? (i.e. 1/4 or 1/3 of the number of photosensors.)

 

 

I'd always thought that when a manufacturer quoted a camera as being 10MP say, there would have to be roughly 3 or 4 times as many actual photosensors, i.e. 30 or 40 million.

Link to comment
Share on other sites

  • Premium Member
Okay, now I'm trying to understand the difference between Bayer and co-site.

 

The way film scanners work is they have line array sensors, one for each color. The film moves past them continuously, and they grab a set of samples each time the electronic clock tells them to. So, the three color samples for a particular point on the film are taken some tiny fraction of a second apart, and put together into a proper three color co-sited pixel in memory.

 

Three chip video cameras use a dichroic prism block to separate the light out into red, green, and blue. Each color gets a chip of its own. The data from the three chips gets assembled into a single co-sited set in memory.

 

Bayer or any other single chip system has only one color per photosite, so they can't produce real sampled co-sited data. They have one real sampled color per location, and they have to figure out what the other two should be by doing maths on the data from nearby locations. This actually works better than in sounds.

 

How many Bayer or vertical stripe photosites equal how many three chip photosites is a matter of substantial controversy and very limited usefulness. Rather than open that can of worms, just shoot Marconi charts. That'll tell you the end to end resolution of the overall system, which is what actually matters to your movie.

 

 

 

-- J.S.

Link to comment
Share on other sites

  • Premium Member

You have to be a bit careful with charts because they tend to involve black lines on a white backdrop. This plays to the strengths of bayer cameras because they see the image with all of their RGB photosites. Try black lines on a red (or more to the point, illuminate the chart with a fairly monochromatic red light source such as an LED array) and things start to fall apart fairly quickly.

 

P

Link to comment
Share on other sites

  • Premium Member
You have to be a bit careful with charts because they tend to involve black lines on a white backdrop. This plays to the strengths of bayer cameras because they see the image with all of their RGB photosites. Try black lines on a red (or more to the point, illuminate the chart with a fairly monochromatic red light source such as an LED array) and things start to fall apart fairly quickly.

 

P

 

Hi Phil,

 

If you did that shooting Red, F35 & 35mm film scanned to 4k what would you expect to see resoloution wise fron the 3?

 

Thanks Stephen

Link to comment
Share on other sites

Hi Phil,

 

If you did that shooting Red, F35 & 35mm film scanned to 4k what would you expect to see resoloution wise fron the 3?

 

Thanks Stephen

 

My argument from above is that all cameras regardless of manufacturer or sensor technology have a performance which is determined by actually measuring the camera system, rather than looking at an image container specification. I wholeheartedly agree that performing good measurements is the only real specification that matters, and yes, you should (and we do) look at resolution across colours. When we do so, we find our cameras actually perform very well indeed.

 

Graeme

Link to comment
Share on other sites

Chaps,

 

Sorry to intrude on the RED compression debate. I read at the top about working in proxies on the edit and then converting to the 4K or whatever resolution you decided to shoot in for colour correction, VFX etc.

 

Excuse my ignorance, however I am very aware of how this operation works and I understand why it works but as of yet I've never worked with R3D footage or proxies for that matter. Can somebody inform me of how to make the proxies in the first place to work with in FCP and then how to re-connect these proxies with the original R3D original resolution footage?

 

Thanks,

 

Jay

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...