Is Red Really 4K?

[richiel]

This is an email I just received after canceling my order for 3 Ikegami HDEditCams and ordering 3 REDS ...any comments? Thanks.........

Just some food for thought -
(Beware of the hype - Ignorance is bliss but knowlege is blissier)

The Red doesn't acutally achieve 4k resolution.
The single CMOS sensor has pixels arranged in a "Bayer" pattern.
It works out to about 60% for Luminance and 30% for color.

The Ikegami however has 3 full res sensors, which gives it full 1920 luminance pixels.

In other words the advantage of the Red it as not nearly as dramatic as they want you to believe.
It is better to the Ikegami but only if shooting in full 4 k mode.
- Except that the Red can shoot 120p some time in the future.
The Ikegami is better in HD mode.
Color separation might be better too. Consider that the Red camera is using Green for luminance.
Perhaps the Ikegami has more true colors.

The other hype thing is editing with Final Cut.
As fas as I gather they do the cut in Final Cut with proxy resolution and finish on a Quantel Pablo.
The first question is why is that a good idea to edit with the raw data stream and extract a proxy from it?
There is overhead in decompressing on the fly and reading a high rate data stream.
It might be much more efficient to create a DV proxy and edit with that.
The other thing is that none of the effects - color correction, special effects, compositing would
translate to the Quantel machine. So all that can be done is just the cut and maybe dissolves
in Final Cut or any other editing system. And all the color correction, effects, compositing etc is
done with the Quantel Pablo.
- something to think about.

 

[Billy Summers]

are you serious...?!

 

[dvpixl]

You've gotta be kidding me.

 

[Tonaci Tran]

alrighty then.

"Hybrid Tapeless Camera GFCAM (Suggested List Price: US$25,000)
2/3 inch, 3 image sensors
HD 1080i/720p format support
MPEG2 LONG/I Frame Multi-codec
4:2:2 digital component recording
MXF file recording
NTSC or PAL MPEG2 recording
Retro Loop function, Time-lapse function, Intelligent Recording
Freeze Mix function
Thumbnail display"

no further comment.

 

[GlennChan]

The other hype thing is editing with Final Cut.
As fas as I gather they do the cut in Final Cut with proxy resolution and finish on a Quantel Pablo.

Because FCP is not a good system for editing 4K. 4K requires fast storage and some heavy-duty processing... which Quantel's custom hardware is good at (and it does background rendering). And Quantel has a much better toolset for online editing (e.g. paint, tracking).

FCP is good at offline editing and isn't designed for 4K online.

The first question is why is that a good idea to edit with the raw data stream and extract a proxy from it?

So you do not need to wait for transcoding during offline. Generating DV proxies would take some time. DV proxies also aren't that high resolution and won't show focus issues; you can of course generate high-resolution proxies like ProRes. But even generating Prores proxies take time.

*Editcam may be somewhat advantageous in its workflow?? (I don't have experience with it so I can't say for sure.)

The Red workflow is designed so that you can capture at very high quality. Offline editorial can ideally be real-time with no transcoding. For finishing the project / onlining, it is slower since you need a step to de-Bayer the material at high quality. This delay is not a big deal in narrative filmmaking.

With the Editcam, you have a (theoretically?) nicer workflow since there is no offline/online... no time is wasted there. Which I presume is very good for news and other quick turnaround markets. You do sacrifice some image quality this way since you are capturing with less efficient compression and the image processing is baked in.

The other thing is that none of the effects - color correction, special effects, compositing would translate to the Quantel machine. So all that can be done is just the cut and maybe dissolves in Final Cut or any other editing system.

Yes this is inherent in the offline/online process. Offline editorial is for getting the story and pacing right. Online editorial is where everything is conformed at high quality, you fulfill your delivery requirements, and where you do some effects and color correction work and other polishing. (*With desktop NLEs nowadays the lines between the two sometimes blur.)

*The Quantel might pick up more effects if you import an AAF (or XML)??? (I don't use a Quantel system so I can't tell you.) Though the effects done at the offline stage is generally to give a rough idea of the effect. They will be finessed and finished properly in the online system... the effect is generally re-done.

Oh, and please don't mind the fanboys.

 

[Mark B.]

It is true that, most of the time, the effective resolution of the RedOne isn't truly 4k due to the Bayer interpolation. But in an ideal situation, if you're filming a picture of a Bayer pattern, then you could hypothetically get true 4k resolution (though if the pattern drifts off center then the resolution could drop to something abysmal). Most of the time though, you'll end up with a resolution somewhere between 1k and 4k. I think someone should do a resolution test for each of the color channels of the RedOne so that we can see those numbers, not just the black and white resolution test.

I don't know how the Red team decided to check for luminance, but they'd have to be crazy not to poll the R & B channels for luminosity in addition to the G channel. If they do poll all the sites, which I suspect they do, then R, B, and G would have surface area coverage of 25%, 25%, and 50%. Ikegami would have surface coverage of 100%, 100%, 100% for both luminance and color. If the RedOne has roughly four times more pixels than Ikegami's camera (4k^2 versus 1920^2), then the RedOne would be 100%, 100%, 200% color resolution in comparison to the Ikegami. If you're shooting 2k scaled, then you're still getting better resolution than the Ikegami. If you're shooting 2k windowed, resolution will drop below the Ikegami in the red and blue channels and subsequently the luminance resolution will be slightly lower as well.

The real benefit of the Bayer is the ability to utilize film-plane camera lenses without special optical adapters. The color and resolution comparisons are less significant because the photo site density of the chips are so large that people will have a hard time discerning the detail of either camera. Both cameras will give excellent picture quality. Build quality and frame rates push the Red past the Ikegami, IMO.

 

[GlennChan]

The Red doesn't acutally achieve 4k resolution.
The single CMOS sensor has pixels arranged in a "Bayer" pattern.
It works out to about 60% for Luminance and 30% for color.

The Ikegami however has 3 full res sensors, which gives it full 1920 luminance pixels.

When you downsample an image, you have to deal with three potential issues:
---Aliasing - the weird moire patterns you sometimes see on fine detail
---Frequency response / MTF - this generally translates to how sharp the image is; though other factors affect perceived sharpness.
---Ringing artifacts - seen as halos and ghosts around objects

If you make the image blurrier, you can get rid of aliasing. This of course makes the image blurrier and affects frequency response. Lenses can make the image blurrier.

With filtering you can improve frequency response (and/or the perception of sharpness). e.g. sharpness settings in cameras
If you oversample the image (e.g. sample at 4K), you can also do certain types of filtering that can get rid of aliasing without making the image blurrier. But both these forms of filtering will introduce ringing artifacts.

1b- You can think of the camera system (lens, sensor design, and signal processing) to be downsampling from the real world, which is (almost) "infinite" resolution to whatever the recording format is.

2a- Oversampling the image (e.g. shooting 4K on the Red) gives you the most flexibility in whatever trade-off you want to make.

2b- The Ikegami camera may also be oversampling the image. If the CCDs are pixel shifted, luma is oversampled and chroma performance is degraded. Most cameras do this since human vision is more sensitive to luma than chroma (in most situations). Also, most cameras apply chroma subsampling, which hurts chroma performance anyways.

2c- Given the same # of pixels, a Bayer pattern is generally more "efficient" than a pixel-shifted 3-chip design. So a 6 megapixel single-chip design will usually outperform a 3 X 2 megapixel chip design.
*Granted, it depends on how you define efficient and there are some tradeoffs between single-chip and 3-chip designs. With 3-chip designs you can't get 35mm DOF without the camera being extremely bulky (AFAIK anyways).

3- Ok, so there are many variables at work and issues to consider. Overall, I would just at the end results that you can get from the camera.

Are 4K Red images downsampled to HD sharp? The answer seems to be yes. *Depends on post processing. You can make the image look a lot sharper by adding contrast and edge enhancement. Most video cameras default to a lot of edge enhancement. Some consider it to be too much, as it leads to halos around edges.
Aliasing: Red cameras have sufficient optical low pass filtering that alias is not terrible. There is significantly less alias with Red images than the majority of 3-chip HD cameras. Lower-quality optics will help HD cameras in this department.
Halos/ringing artifacts: You can tweak this to taste.
DOF: Red has 35mm DOF, unless shooting in the windowed modes. Sometimes deep DOF is advantageous (news).
Compression artifacts: Red should be better than any chroma-subsampled compression format (e.g. DNxHD). At very high bitrates/quality, chroma subsampling hurts the image IMO.
De-mosaic artifacts: Very little/none.

Noise, sensitivity: The sensor is very clean. Take a look at the test chart shots and compare them with other cameras. The compression is also very clean.
Performance with greenscreen: The high resolution and low noise really excel here. The Red should be very nice for greenscreen work.

Visual quality: This depends on your taste. But the images look good in my book. *It also depends on what color grading you do. Film has its own aesthetic/look.
Workflow: It depends on what you do / what systems you use. The Editcam should have a superior workflow for NLEs that support it.

 

[I Bloom]

It's a common misconception...

It's a common misconception...

I want to challenge the conception that a 4K Bayer pattern image is not 4K. It's not true and it's based on a misconception of how bayer interpolation works.

Here is what I believe to be a more true statement:

A 4K Bayer Pattern image is 4K but it contains artifacts because the interpolation process is imperfect.

This is really a different thing than saying that the image does not have full resolution. If you believe a 4K Bayer pattern is really 1K than would you shoot with a 1K RGB camera and expect it to be as sharp as a 4K Bayer pattern... I don't think you would.

Simply dividing out the samples per channel ignores the underlieing principle behind the Bayer pattern, which is that we can use pixels from one channel to predict the other channels based on the color ratio in any given area. After that we can use other algorythms to improve the overall quality of the image. We simply can't do this when we interpolate RGB to a larger size.

Most images that we will find in nature the intensity of light in a detailed area will change more rapidly than the color ratio of each pixel. In images where the color ratio changes gradually Bayer pattern interpolation can approach what would have been recorded were the image RGB.

In other words we are exploiting natural properties of real world images in such a way that we can forget 2/3's of the information.

The truth is, that under most conditions a 4K Bayer pattern interpolated with an advanced algorithm can appear as sharp as a 4K RGB. And this appearance is what matters.

IBloom

 

[Robert Berger]

Ikegami

Ikegami

I want to challenge the conception that a 4K Bayer pattern image is not 4K. It's not true and it's based on a misconception of how bayer interpolation works.

Here is what I believe to be a more true statement:

A 4K Bayer Pattern image is 4K but it contains artifacts because the interpolation process is imperfect.

This is really a different thing than saying that the image does not have full resolution. If you believe a 4K Bayer pattern is really 1K than would you shoot with a 1K RGB camera and expect it to be as sharp as a 4K Bayer pattern... of course not.

Simply dividing out the samples per channel ignores the underlieing principle behind the Bayer pattern, which is that we can use pixels from one channel to predict the other channels based on the color ratio in any given area.

Most images that we will find in nature the intensity of light in a detailed area will change more rapidly than the color ratio of each pixel. In images where the color ratio changes gradually Bayer pattern interpolation can approach what would have been recorded were the image RGB.

In other words we are exploiting natural properties of real world images in such a way that we can forget 2/3's of the information.

The truth is, that under most conditions a 4K Bayer pattern interpolated with an advanced algorithm can appear as sharp as a 4K RGB. And this appearance is what matters.

IBloom

Whats the problem!
\I have been testing the Ikegami HD camera, it doesn't look a bit of the picture i saw at the IBC from the REd One. I don't care if the real resolution is 4K before or after FC. I just look at the picture and it looks beter than 4K showed films at the 4K projection show from Sony. The film grain moved through the images from spiderman and the red movie was clean en smooth.
And don't forget the man behind the camera make the shots.......

with kind regards
Robert

 

[Graeme Nattress]

"Color separation might be better too. Consider that the Red camera is using Green for luminance." this is just so untrue. It shows a complete lack of understanding of what we're doing.

As for image, Any camera with 1080p sensors cannot produce a 1080p image without aliasing any more than a 4k sensor can produce a 4k image without aliasing, and you definately don't want bags of nasty aliasing.

Ikegami cameras have a prism. Seen any prism artifacts lately - yuck.

They have 2/3" sensors - so you can't use cinema glass, can't get the right DOF characteristics.

Oh, and listen to Glenn.

Graeme

 

[Lauri Kettunen]

Not sure if there's a point to put the spoon on this soup, but I think using the number of pixels to compare a CMOS sensor camera to a 3 sensor one is bit like evaluating cars and motorcycles based on the number of wheels. The numbers should and cannot be employed to evaluate all aspects.

Probably, for many camera owners one of the main issue is, which system yields the best image quality for a given amount of money -lenses included. From the electronic point of view a 3 sensor camera is a brilliant idea. The drawback is hidden in the prism and what it implies in lens design. High quality lenses for 3-sensor cameras tend to be expensive. Cameras with CMOS sensors bypass this pitfall.

One important detail is that both a 3-sensor and a CMOS-sensor need a low pass filter in front of the system to avoid aliasing. In simple, but non-accurate words, a low pass filter is like unsharpening, i.e. small details are filtered out , i.e. "blurred". This is important, for if you even you hypothetically shoot the negative of the bayern pattern with perfect aligment, due to this filtering the neighbouring pixels will get some information of the light rays ending otherwise only in the actual pixel. In this sense, one may argue that since the low pass filtering is a must, the CMOS sensor turns this unavoidable necessity to somewhat a bonus. This is a reason, why in the debayerian process one may save more data than what one may think at the first glance. (I should thank Graeme for explaining this point of filtering some time ago.)

In the end of the day, it seems that currently a CMOS sensor camera like RED has the highest price/quality ratio.

 

[Graeme Nattress]

I don't think you could seriously call any camera 3 chip, single chip or otherwise a good engineering design, without some form of optical low pass filtering. To try to avoid it really says you believe in marketing more than sampling theory. Practical optical filters put limits on resolution extraction and aliasing. Within those limits the Bayer pattern is very efficient and cost effective, and allows you to avoid the prism. All in all, it's probably the best balance in engineering design for now.

Graeme

 

[GlennChan]

To add onto what Graeme is saying...

Some links, resources for understanding sampling (in non-technical terms):
http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm
http://www.cambridgeincolour.com/tutorials/digital-photo-enlargement.htm

You can grab a zone plate image off here:
http://www.worldserver.com/turk/opensource/index.html#ZonePlate
You can then bring it into an (image) editing application and play around with it there. Try different resizing algorithms and try blurring the image beforehand.

Also make a text image and resize that. This will let you see any ringing artifacts (some but not all algorithms introduce them).

And also try real world images (esp. moving ones). Aliasing isn't always as nasty as what the zone plate would suggest. But it can and does happen. e.g. usually you see it on slow pans across buildings... the line has a 'crawling' movement on the edge. Certain fabrics and patterns are also prone to aliasing (ones with lots of fine detail).

2- As far as sharpness goes, see
http://www.cambridgeincolour.com/tutorials/sharpness.htm

3- If you want the explanation of sampling theory in technical/engineering terms, try Charles Poynton's book "Digital Video and HDTV".

 

[georged]

Do we need 4k workflow?

Do we need 4k workflow?

Pretty confusing... but important to really get to the ground of this 4k resolution / Bayer pattern issue.
What is the benefit finishing in 4K?
It seems that a 2k image would hold just about the same amount of detail.
Considering the enourmos cost in time and $ of 4k editing - would it make more sense to finish in 2K?

 

[Joe Carney]

This is an email I just received after canceling my order for 3 Ikegami HDEditCams and ordering 3 REDS ...any comments? Thanks.........

Just some food for thought -
(Beware of the hype - Ignorance is bliss but knowlege is blissier)

The Red doesn't acutally achieve 4k resolution.
The single CMOS sensor has pixels arranged in a "Bayer" pattern.
It works out to about 60% for Luminance and 30% for color.

The Ikegami however has 3 full res sensors, which gives it full 1920 luminance pixels.

In other words the advantage of the Red it as not nearly as dramatic as they want you to believe.
It is better to the Ikegami but only if shooting in full 4 k mode.
- Except that the Red can shoot 120p some time in the future.
The Ikegami is better in HD mode.
Color separation might be better too. Consider that the Red camera is using Green for luminance.
Perhaps the Ikegami has more true colors.

The other hype thing is editing with Final Cut.
As fas as I gather they do the cut in Final Cut with proxy resolution and finish on a Quantel Pablo.
The first question is why is that a good idea to edit with the raw data stream and extract a proxy from it?
There is overhead in decompressing on the fly and reading a high rate data stream.
It might be much more efficient to create a DV proxy and edit with that.
The other thing is that none of the effects - color correction, special effects, compositing would
translate to the Quantel machine. So all that can be done is just the cut and maybe dissolves
in Final Cut or any other editing system. And all the color correction, effects, compositing etc is
done with the Quantel Pablo.
- something to think about.

based on the descripton of their own camera (1080i) a much better price/performance camera is the new Sony XDCAM XE. 3 True 1920x1080p chips, uncompressed sdi out...for less than 1/3 the price. Who needs Ikegami at all? just food for thought.

 

[I Bloom]

Pretty confusing... but important to really get to the ground of this 4k resolution / Bayer pattern issue.
What is the benefit finishing in 4K?
It seems that a 2k image would hold just about the same amount of detail.
Considering the enourmos cost in time and $ of 4k editing - would it make more sense to finish in 2K?

I think you might be overestimating the cost of finishing in 4K. Consider for a moment that REDCODE compression is around 10:1. So the amount of data in the pipeline is actually managable.

In a similar way the codec make it possible to work with REDCODE without fully decompressing the image. Thus its possible to cut an entire RED piece in Final Cut Pro as say a 1K movie, and then start working in 4K only during the very last stages of finishing.

Exactly how much computing power is needed to do this is unclear, but it seems likely that a fairly standard modern Final Cut Studio setup will be sufficient and possibly even a Macbook pro laptops. High end solutions might only come in to play at the very end, if at all.

And also, RED should at some point be offering 2K windowed RAW and 2K scaled RAW as options, in those cases that 4K is too much. http://www.redhax.net/wiki/Camera_Formats

IBloom

 

[Graeme Nattress]

2k does not hold the same amount of detail as the 4k image. Uprezzing the 2k to 4k shows it to be inferior.

4k is not necessarily that much more expensive than 2k.

Graeme

 

[M Most]

2
4k is not necessarily that much more expensive than 2k.

That depends on what you're planning to deliver. You have thought about deliverables, haven't you?

 

[laguun]

nothing against ikegami, the editcam is a nice camera, however just some things to consider.

* editcam is a eng/efp camera. its main strenght is its open workflow (edit on camera while driving back from the shooting), edit on a large selection of pcs nles running windows & osx and writing directly to several codecs.
* another big advantage: if you buy editcam today, they will deliver quickly, ordering reds will require some months.
* they can write several codecs, which is helpful in many scenarios.

red strenghts are much more in the creative side:
* sensorsize : s35mm, s16mm, 2/3 DOF - ikegami has only 2/3
* framerate : 1-100p, allowing much better slowmo, important for drama
* single chip sensor, allows the use of cooke, angenieux, zeiss/arri cineglass, nikon and canon if you want to be highly economic with the lens investment and have a high quality. red cineglass as mix.
* 4k resolution. No matter how you turn it, a 12MP pixel sensor after debayering resolves much more than a 3*2K configuration. Almost -any- manufacturer in moving images (arri, red, dalsa, si...) and in still images (hasselblad, canon, nikon...) use debayered CMOS in its present top-end products today.
* datarate. red is at 270Mbit, ikegami at 145 (will be 220 sometimes)
* price. the red is a great offer. editcam hd is rather average.

Furthermore, the ikegami sales person makes some mistakes in its claims.
when he talks about debayering, he should mention that the editcam is 4:2:2 YUV. So this means: 1920 luma, 960 chroma resolution is maximum for editcam.
Then, the colordepth. editcam has only 8bit. Having mastered and colored many many full features and commercials i can tell you that 8bit are hard to deal with. Its fully possible, buts its much easier with 10bit.

As every camera, red has its limits as well. As of now, there are still some important features missing in the camera´s firmware. However, they will be added in the near future, certainly before you get delivery of your reds if you order now.

In one thing ikegami is correct: red doesn´t supported 4k10bit RGB editing, but only 2k8bit RGB or 2KfloatYUV editing as of now, which is limited by Apple Final cut pro. The realtime 4k editing/finishing systems (as iQ/Pablo or dvs clipster) and 4k10bit RGB softwares (as Adobe Premiere) aren´t supported as of yet.

So bottomline: With the RED you get a much more versatile, more flexible and creative tool which shooter much better images for a lower price. However you have to wait for delivery and will have no inexpensive 2k10bit rgb or 4k proxy workflow.
With the ikegami, you can get the camera quick, you will have several very fast workflows (avid, apple, adobe etc supported) and a good imagequality. However you will pay a much higher price for less imagequality, less creative option and will probably have to invest in pretty expensive 2/3 HD lenses as well.

I hope this was balanced enough. If you are more in ENG and think about the editcam, i also would consider having a look at the GV Infinity, now, as they finally started shipping. We will skip editcam/infinity/xdcam hd/p2 and are pretty happy with our combined hdcam & red gear.

 

[mezmo]

I think you might be overestimating the cost of finishing in 4K. Consider for a moment that REDCODE compression is around 10:1. So the amount of data in the pipeline is actually managable.

In a similar way the codec make it possible to work with REDCODE without fully decompressing the image. Thus its possible to cut an entire RED piece in Final Cut Pro as say a 1K movie, and then start working in 4K only during the very last stages of finishing.

Exactly how much computing power is needed to do this is unclear, but it seems likely that a fairly standard modern Final Cut Studio setup will be sufficient and possibly even a Macbook pro laptops. High end solutions might only come in to play at the very end, if at all.

And also, RED should at some point be offering 2K windowed RAW and 2K scaled RAW as options, in those cases that 4K is too much. http://www.redhax.net/wiki/Camera_Formats

IBloom

Hi IBloom,
So you will have to render out at some point a 4K DPX, or other
4K flavoured file for final finish of the project. How do you plan on doing
this and on what hardware?
Cheers Mezmo