Peter Moretti

"PIXEL SHIFTING AND UP-REZZING NOT SPOKEN HERE" is stated on Red's website. But it that really accurate?

The camera uses a single Bayer Pattern sensor. A Bayer sensor records color in a 4:2:2 ratio by virtue of the fact that it's 50% green pixels, 25% red ones and 25% blue ones. Yet Red produces 4:4:4 color. But FWI, those last two 4's must come as a result of demosaicing the Bayer patter. So there IS image processing going on.

I'm not saying that the Red sensor doesn't create a superb image, I'm sure it does. I'm just saying that, FWIU, some pretty heavy math, in the same league of pixel shifting and up-rezzing, is being applied to the image.

Mathieu Ghekiere

I dont know much about those very technical things, but RED produces RAW, not 4:4:4, although you can process it to a 4:4:4 file.

__________________
http://productionsofthe3rdkind.com
http://pr3k.com

Peter Moretti

But the RAW file is inherently subsampled. I'm not saying that the results aren't excellent. But to get 4:4:4 color out of a Red, image processing has to be applied to the RAW footage. And FWIU, the demosaicing needed with a Bayer sensor is more extreme than the 1/2 pixel shift that many cameras use.

David Heath

Terminology such as 4:4:4 etc really applies to the recording method, and doesn't have a direct meaning when applied to sensors etc. 4k 4:4:4 implies by defintion a 4k matrix with all colours having full and equal resolution. A 4K Bayer sensor won't deliver that. What it will deliver is very approximately 70% of 4k luminance resolution, (depending on algorithm, scene content etc) and somewhat less chrominance resolution. Recording it in 4k 4:4:4 form does not mean it will look as good as an image from theoretical 3xRGB 4k sensors - anymore than recording the output from a very cheap camera will be improved just because it was recorded to HDCAM SR, say.

None of this is to denegrate the RED approach - quite the opposite - but to try to distinguish between actual resolution and recording format. The resolution of the latter only defines the maximum possible, not what it actually is.

All that really should be said of Red in this respect is that "it's a 4k Bayer sensor".

Peter Moretti

And isn't it true that to use the image off of a Bayer sensor, the signal always has to be demosaiced or deBayered? This can be done right off the block or after recording, but it always has to be performed, even with a Red, correct?

So while pixel shifting isn't being done (it can't be with a single sensor camera) demosaicing always is.

I believe this is the case with ALL single sensor cameras, even the Dalsa Origin, so I'm not trying to single out Red.

Thomas Richter

To me it feels like the 3k bayer had been chosen to get an output similar to a 2k film scan.

Would explain why Epic is 5k - the attempt to come closer to the resolution of a 4k film scan

Seems Red paid attention to the resolution chart results.

Just my 0.02$

Juan M. M. Fiebelkorn

This takes me to the point of wondering: What is the great difference between Demosaicing a Bayer Pattern or "demosaicing" a pixel shifted image? :)

Ali Husain

probably you can get better colors out of pixel shifting and multiple imagers with fancier color filters (compared to the micro color filter array on CMOS sensors).

Steven White

Chroma sub-sampling notation is really not appropriate for a RAW image off a bayer patterned sensor.

When trying to consider the possible resolution of a debayered image off a 4K sensor, you have to remember that 4K simply refers to the number of photo sites. Since the photo sites do not have 100% coverage, Red employs an optical low pass filter (OLPF) to effectively blur the incident light a little.

Debayering is not done online in a Red camera, because on-camera algorithms are more limited due to their real-time requirements. Instead, Red does debayering in post, taking the 12-bit linear RAW data, and calculating a full 4K RGB image (I recall with 16-bit per colour precision).

The effective resolution of the image must then be measured from the image. Due to the OLPF and the debayering process, the maximum effective resolution of Red images is usually estimated around 3.2K or so.

It is reasonable to expect that the final resolution from Red compares to that of a 12 mega-pixel bayer patterned full frame digital SLR.

All that said, most of the time when you see a "soft" image from a Red One, it's because the lens wasn't focused correctly.

3-sensor pixel shifted arrangements aren't usually any better, since they have to deal with aberration, also need an OLPF, and require all three chips to be perfectly aligned. Issues with any of the above three requirements decrease the achievable resolution of the final image.

Glenn Chan

They are using Bayer pattern, which is not the same thing as pixel shifting. In that sense it's true.

The other thing they are getting at (as I interpret it) is that the Bayer pattern is more efficient than pixel shifting. 6 megapixels in Bayer arrangement is more desirable than 3X2 megapixels in a pixel shifted 3-chip arrangement. Both arrangements optimize for luminance resolution over color performance. Since our visual system has higher luminance acuity than color acuity, it's mostly ok to do that / it's the least bad compromise.
If you look at the color cones in the eye, the eye is Bayer-ish in that there are a lot of green-sensitive cones compared to red and blue sensitive ones. (The real answer is a lot more complicated since the % of blue cones suggests that we could use a lot less blue resolution, but this is not the case. I don't think we fully understand the human visual system yet.)

I don't think 4K film scans really have that much resolution.

I believe the point of the 5K is that you have a little more resolution once you go down to 4K. I haven't played with this stuff myself though.

Glenn Chan

I wouldn't use chroma subsampling notation to describe the sensor sampling.

A- Chroma subsampling (in current practice) is applied to gamma corrected signals, which will cause shifts in luminance. The images from a Bayer sensor doesn't necessarily have to suffer from this (Red handles this correctly).

B- Chroma subsampling can cause artifacts where the subsampling effectively creates negative or out-of-gamut colors. The resulting values call for negative light, which doesn't exist. Bayer sensors don't inherently record negative light (because it doesn't exist).

Basically, chroma subsampling and Bayer sampling is not the same process.

Ethan Cooper

Have you seen the images that come out of that camera body that costs $18,000? I still don't understand why people try to rip this camera.

__________________
The large print giveth, the small print taketh away.
-Ethan Cooper www.silver-media.net

Jon Fairhurst

In one case a company puts out a camera with essentially a PAL SD sensor, pixel shifts and tells us it's an HD camera.

In this case, RED puts out a camera with 3K pixels. Rather than claim that they have made a 4K camera due to pixel shifting, we realize that this will probably yield a true 2K to 2.5K result - in a <$3,000 camera! This rocks!

Their first product has 4K+ pixels, and only later did people (including themselves) realize that the effective result probably isn't quite 4K. So they address the issue by making a 5K camera. Again, this rocks!

We're getting the real story in both real and effective resolution. I'm cool with that...

__________________
Jon Fairhurst
FRESH FILMS WEEKLY AT COLONELCRUSH.COM.

Chris Hurd

That's because it *is* an HD camera. Remember what matters is the output, not what happens at the sensors. Remember the exact same scenario happened before with the original Canon XL1. It had sub-SD sensors with Pixel Shift. Nobody complained about it then.

Also keep in in mind that with very few exceptions, *every* three-chip camera design incorporates some form of pixel offset resolution boost, whether it's Pixel Shift, ClearVid, or whatever. JVC ProHD being the one notable exception.

It's a serious mistake to focus only on the resolution of the sensor because that's not the only component in the chain affecting the final output. You can't ignore or dismiss Pixel Shift because it's a critical step in the process.

Kudos to RED's marketing strategy for playing the "Pixel Shift not spoken here" card; they've done a good job of selling the difference. Too bad they're feeding the myth that Pixel Shift is somehow a "bad thing."

__________________
CH

Search DV Info Net for quick answers | The best in the business: DV Info Net Sponsors | ...Tuesday is Soylent Green Day!

Ethan Cooper

720p, 1080i, 1080p, 2K, 3K, 4K, 5K, HDV, DVCProHD, HDCAM, RAW, it doesn't matter. Does the camera in question produce visually stunning images? Is the platform scalable from run & gun indies to regional and national broadcast, to feature films? In Red's case the answer is yes. All other debates are secondary in my estimation.

__________________
The large print giveth, the small print taketh away.
-Ethan Cooper www.silver-media.net