Other 4K digital cinema cameras

[Tom Lowe]

Keep in mind that most "4K" material in 4K reels probably wasn't even acquired at 4K. For that, you would need a 6K Bayer camera, and the only game in town right now would be timelapse shot on 21MP or 24mp RAW DSLRs.

 

[Stuart English]

4K material

4K material

I bet it was because:

a 4K Bayer single sensor camera is a 4K camera, just as a 4K RGB 3-chip or 4-chip + prism based camera is too (if any exist)

Its true neither one will resolve a full 4,096 lines - as Graeme has explained - but they would both certainly be capturing "at 4K"

 

[Bill Anderson]

It's often said, in the discussions about the desire for larger
sensors, that the eye, or projector, or intermediate film stock,
or TV, and so on and so on, is only capable of resolving x amount
at y distance.... I think most of us might agree with many of
the claims about limited returns in advances in resolution if
resolution were only that. But hard discernible information
is not the only game in town as far as resolution goes.
What greater and greater resolution brings to us
(in the shape of larger formats) is often manifested in less
tangible or objective ways: luminosity, color fidelity, pearl
like whites rolling effortlessly down the subtlest scale to
reach satin blacks that can offer up a noiseless hint of what
they conceal. Color blending from one end of the spectrum
to the other that will compete with the seamless logic of
an evening sky. An image that quietly invokes a depth
reminiscent of a window, rather than a flat screen.
You know, all that romantic artsy subjective stuff. The stuff
that allows you a bigger room to swing your schtick in.
We're at a fine place now, but it's going to get finer. I can't wait [ :

 

[Jerrod Cordell]

On a closely related note, in my experience, 4K digitally projected images that have no gate weave and minimal flicker will reveal the massive amount of motion blur in 24fps material. I estimate that 60fps at 4K will provide the "looking out the window" experience, can't wait to see if I'm right. (Yes, I realize that it might not "feel cinematic")

Yuck. As long as it doesn't look like the 120 hz HDTVs. I can't stand it. It makes Lost look like it was shot on a very high quality handicam.

 

[M Most]

What greater and greater resolution brings to us
(in the shape of larger formats) is often manifested in less
tangible or objective ways: luminosity, color fidelity, pearl
like whites rolling effortlessly down the subtlest scale to
reach satin blacks that can offer up a noiseless hint of what
they conceal. Color blending from one end of the spectrum
to the other that will compete with the seamless logic of
an evening sky... [ :

Those are all very nice words, but what gets you those things is not more resolution, but more bit depth. Smooth color blending is achieved by a closer approximation of the analog original, which has, in effect, infinite bit depth. The more bits you use, the closer your digital version gets. Given a screen size, resolution ultimately gets to the point of diminishing returns. The ultimate fidelity of an image is a combination of many factors, as well as the size of the screen being used.

 

[Graeme Nattress]

No, analogue does not have equivalent to infinite bit depth. In an analogue signal, the equivalent to bit depth is noise. The equivalent to resolution is bandwidth.

There are benefits to higher resolution other than perceivable detail though - the oversampling benefits we talked about earlier. They can be perceived as the improvements Bill mentions.

Graeme

 

[M Most]

No, analogue does not have equivalent to infinite bit depth. In an analogue signal, the equivalent to bit depth is noise. The equivalent to resolution is bandwidth.

I wasn't necessarily talking about electronics. I was talking about the real world and how the human eye perceives that world. And in that context, I stand by my statement.

There are benefits to higher resolution other than perceivable detail though - the oversampling benefits we talked about earlier. They can be perceived as the improvements Bill mentions.

While that's somewhat true, I didn't interpret the original statement as being related to oversampling. Maybe I misinterpreted.

 

[Yousuf Abbasi]

In an analogue signal, the equivalent to bit depth is noise.

Noise... Really? In what sense are you referring to "noise?" In film, it's grain that's comparable to digital "noise." Bit depth relates to the gradient of color, so I don't see the connection to noise. I concur with M Most, in that analog is equivalent to infinite bit-depth.

 

[Blair S. Paulsen]

Don't hate me because I'm a threadjacker...

Don't hate me because I'm a threadjacker...

Assuming that well engineered compression allows us to work with greater resolution and bit depth on available, non-esoteric hardware (somewhere between desktop and Pablo) then the more the merrier. It will be interesting to test the higher bandwidths expected with the EPIC and see where the tradeoffs lie between amount of compression and ease/cost of use.

Shawn's observations of RC28 v RC42 may be indicative of the real world limitations of viewing systems or simply that other factors play a much greater role than bandwidth at this particular point in the evolution of the entire system. Jim has indicated that the only significant motivation for pushing up throughput is to get higher resolution when overcranking so perhaps those of us that have traditionally used compression ratios as a quality metric need to re-calibrate our mindsets.

Do we know if the bit depth of RedCode variants to come will remain at 12 bit linear? Not to put words in his mouth, but Graeme has indicated that more bit depth than 12 is mostly noise. I assume that's based on the inherent limitations of the sensors currently available, primarily S/N ratio.

My guess would be that the Monstro and other next gen sensors will make the upside of 14 or 16 bits more attractive despite the bandwidth penalty. I am hoping that the prowess of modern codecs like RedCode will make this equation pan out. I am making a further assumption that one of the other goals of the EPIC program and the RedCode variants in the 100MB/sec or higher range are not just to reduce the compression ratio and/or allow higher frame rates - NOT that there's anything wrong with that - but also to blaze the trail for greater bit depth implementations when the noise floor of the chip justifies it.

Reportedly the new Arri D cams will offer a 16 bit output, presumably in ARRIRAW and if I read it right, downsampled from 3 or 3.5K to 2K in camera using a dual gain path topology that sounds interesting.

Perhaps in 5 years or less we will have 16bit, 15 stop DR, ISO 1600 acquisition that will make grading more about selectively tossing the stops of DR that can't be put into the deliverable container than massaging the data set to control noise, limit artifacting or expand latitude. :grouphug:

Cheers - #19

 

[XiaoSu Han]

I wonder why we are trying to achieve more resolution?

What's the rational reason behind it besides "bigger, better and sharper"?

What's the rational reason behind having 8K over 4K?

I see why we'd love to have a 1litre car instead of a 4litre one, but why 8K over 4K?

Wouldn't rates at for example post houses become cheaper if they would invest in a 4K suite today and not have to upgrade to 8K in two years? Wouldn't the 4K suite be paid off quite soon and more money could go to the operator and his skills instead of un-necessary technology. For example I can't really take the really sharp 1080p real HD image nowadays anymore even, how am I supposed to endure 8K? And what for? We already have a great image projected in IMAX theatres nowadays, what should improve?

Some people say it's important if you get close, but I have my favorite distance to the screen already so why would I even want to go close to the screen?

Maybe tastes are different as are affections

 

[Steven Caesare]

Noise... Really? In what sense are you referring to "noise?" In film, it's grain that's comparable to digital "noise." Bit depth relates to the gradient of color, so I don't see the connection to noise. I concur with M Most, in that analog is equivalent to infinite bit-depth.

Theoretically, yes. Practically, no.

All circuts, including analog, have some noise present in them, from power supplies, thermal fluctuation, solar radiation, electromagnetic interference eminating from deep space, etc...

In an analog system, this noise effectively limits the resolution of the waveform. If the noise averages ~0.001 volts, and you have a 5v peak-to-peak working voltage in your analog circuitry, you cannot exceed about 5000 steps of reolution. Incidentally, that happens to be about 12 linear bits.

This is why my analog audio mixer uses ultra-low impedance circutry... it reduces the induced noise in the components. But it doesn't have infinite resolution.

Once you have sufficient bit-depth to capture the dynamic range of the signal without discernable steps (gradients in video, audible volume "steps" in audio), then more doesn't really buy you anything in the capture/storage stages (but can be helpful during manipulation).

Interestingly, while "24bit" audio has been all the rage for audiophiles, and in the new HD disc formats, there have been quite a few test indicating that even good audio gear can't reproduce much more than about 14-15 bits of audio fidelity, and GREAT gear will often hit only 19 bits or so. Noise is often the limiting factor... to the extent that some high-end gear goes so far as to to allow you to turn off the electro-luminescent front panel display, as the power supplies to drive those introduce enough voltage fluctuation in the power rails that it makes a difference performance of system.

-sc

 

[Steven Caesare]

Ass

SNIP

Do we know if the bit depth of RedCode variants to come will remain at 12 bit linear? Not to put words in his mouth, but Graeme has indicated that more bit depth than 12 is mostly noise. I assume that's based on the inherent limitations of the sensors currently available, primarily S/N ratio.

My guess would be that the Monstro and other next gen sensors will make the upside of 14 or 16 bits more attractive despite the bandwidth penalty.


SNIP


Cheers - #19

New sensors with 14 bit A/D stages are already on RED's announced roadmap.

-sc

 

[Graeme Nattress]

Steven, that's a pretty good explanation on bit depth! In a digital system, there is quantization noise or quantization error from the size of the quantization or bit depth. As soon as your quantization noise is lower than the noise in the analogue signal, there is no benefit to increase the bit depth any further.

No signal of any kind is ever free of noise, and that is why noise in an analogue signal limits it's information content and is absolutely equivalent to bit depth in a digital signal. Bandwidth is equivalent to resolution (or sample rate).

Graeme

 

[Charles Angus]

You could argue that But fact is, it doesn't measure anywhere remotely near that resolution. I'd still like to point one at my zone plate though, to see what it actually does measure.

There's also the big NHK 8k R&D camera that gets shown at NAB.

Graeme

I bet if you could take it to Panavision, they'd let you...

 

[Steven Caesare]

Steven, that's a pretty good explanation on bit depth! In a digital system, there is quantization noise or quantization error from the size of the quantization or bit depth. As soon as your quantization noise is lower than the noise in the analogue signal, there is no benefit to increase the bit depth any further.

No signal of any kind is ever free of noise, and that is why noise in an analogue signal limits it's information content and is absolutely equivalent to bit depth in a digital signal. Bandwidth is equivalent to resolution (or sample rate).

Graeme

Indeed.

I think a confusing factor for many (including some posters I've seen heres) is that bitdepth is used to indicate dynamic range of a signal (ie 12 bits to capture ~12 stops), as well as sample (not spatial) resolution (ie 8 bit color, etc...).

The truth encompasses both. While you can have infinitely large DR in just one bit (i.e. a light switch that's on or off), it takes a sufficient number of bits in a sample to capture the DR of the input with a sufficent number discrete values such that the steps between the values are not discernable (by eye or ear).

Thus if you increase the DR of the input (ie, the Monstro sensor @ ~14 stops), while you COULD capture that in 12 linear bits (or less), you'd likely get banding, therefore 14-16 would likely be a better choice.

However, while jumping to 20 bits would effectively give you a greater sample resolution, it wouldn't do much good, as you would only be taking steps that you already can't see at 14 bits, and dividing them in to smaller steps you also can't see.

Furthermore, if the noise in the system is obscuring the last 2-3 bits, then you are limited anyway.

I of course know you know all of this Graeme, it just seems that bit depth, sample resolution, noise floor, dynamic range, and the like cross some folks here up.

-sc

 

[Michael Vera]

Theoretically, yes. Practically, no.

All circuts, including analog, have some noise present in them, from power supplies, thermal fluctuation, solar radiation, electromagnetic interference eminating from deep space, etc...

In an analog system, this noise effectively limits the resolution of the waveform. If the noise averages ~0.001 volts, and you have a 5v peak-to-peak working voltage in your analog circuitry, you cannot exceed about 5000 steps of reolution. Incidentally, that happens to be about 12 linear bits.

This is why my analog audio mixer uses ultra-low impedance circutry... it reduces the induced noise in the components. But it doesn't have infinite resolution.

Once you have sufficient bit-depth to capture the dynamic range of the signal without discernable steps (gradients in video, audible volume "steps" in audio), then more doesn't really buy you anything in the capture/storage stages (but can be helpful during manipulation).

Interestingly, while "24bit" audio has been all the rage for audiophiles, and in the new HD disc formats, there have been quite a few test indicating that even good audio gear can't reproduce much more than about 14-15 bits of audio fidelity, and GREAT gear will often hit only 19 bits or so. Noise is often the limiting factor... to the extent that some high-end gear goes so far as to to allow you to turn off the electro-luminescent front panel display, as the power supplies to drive those introduce enough voltage fluctuation in the power rails that it makes a difference performance of system.

-sc

So would "infinite resolution" in analog circuity, be the equivalent to saying "infinite bandwidth?"

And I agree, the inherent noise in everything really does limit the higher end gear, just wont get the benefit of the maximum 24 bits

 

[M Most]

I of course know you know all of this Graeme, it just seems that bit depth, sample resolution, noise floor, dynamic range, and the like cross some folks here up.

Other things cross them up even more, such as when you consider perceptual factors. That's why 10 bit log coding can contain virtually all of the perceptible information in a 12 bit linear light original image. When you're talking about mathematical processes, you can come up with absolutes. When you take into consideration perceptual factors, the rules become a little more blurred and prone to subjective opinion.

 

[Steven Caesare]

So would "infinite resolution" in analog circuity, be the equivalent to saying "infinite bandwidth?"

And I agree, the inherent noise in everything really does limit the higher end gear, just wont get the benefit of the maximum 24 bits

I was just speaking to a close friend about this. :wink5:

It would. Excpet that you said "circuitry", of which there are none that can do this in practice.

It's one of those theory vs. reality things.

-sc

 

[Steven Caesare]

Those are all very nice words, but what gets you those things is not more resolution, but more bit depth. Smooth color blending is achieved by a closer approximation of the analog original, which has, in effect, infinite bit depth. The more bits you use, the closer your digital version gets. Given a screen size, resolution ultimately gets to the point of diminishing returns. The ultimate fidelity of an image is a combination of many factors, as well as the size of the screen being used.

Your own example of spatial resolution being of diminishing value beyond a certain point holds for bitdepth (ie- discrete values represeting a color) as well.

Beyond the point where you can no longer detect the graduation form one color to the next, you have exceeded your eyes ability to resolve anything greater. More bit depth there will not help.

Thus your last statment: "The ultimate fidelity of an image is a combination of many factors...", is somwhat at odds with your leading statement: "Those are all very nice words, but what gets you those things is not more resolution, but more bit depth.".

You had it right the second time... it's the combo.

 

[pattinsonrobert]

Hi Everyone..

If you shot at 4K 16x9, then Log and Transfer will bring in the footage at 2048x1152, so set up your sequence for that.

If you shot 4K HD, Log and Transfer should bring it in at 1920x1080. I haven't tried it though, but that's what it should do.

Regards
Pattinson.