Trading 3K (4K) for 2K (3K) with better DR, less noise and other improvements?

[Peter Majtan]

I am creating this poll in response to many people bashing the idea that 4K (and 3K in case of Scarlet) is little overkill and would rather trade it in for better dynamic range, less noise and other benefits...

The main assumption being that 2K is more then enough for many of us.

Having less pixels in the same size sensor gives one much less noise and improves dynamic range. It also allows for faster frame rates and/or less compression within a specific capture bandwidth (100MBps). There are many more benefits and the only down-side is the less resolution - but does size really matter too much?

This question is even more relevant for other countries where HD is still some years away from becoming a broadcast standard. Also consider that majority DI post is being done in 2K or HD. Including STAR WARS!!!

I am even proposing a new camera between Scarlet and Red One (more on this here: http://www.reduser.net/forum/showthread.php?t=12235 ) which could take advantage of this fact.

Please vote in the poll (note the double options for people in USA and outside - no discrimination intended, just trying to see the difference in preferences...) and share Your opinions in this thread...

And please keep it chilled... :innocent:

 

[Chris Kenny]

Remember, bayer pattern sensors only resolve about 75% of what their raw photo site count indicates. A 3K sensor is going to produce noticeably better 1080p HD than a 2K sensor would. I think Red has made exactly the right choice here.

Note that having fewer pixels on the sensor results in less noise and better dynamic range (same thing, actually) on a pixel-for-pixel basis. But downscaling also results in less noise and greater dynamic range on a pixel-for-pixel basis, plus you still get a sharper image. This doesn't quite make up for the smaller photo sites, because of fill-factor issues, but it is significant.

 

[Peter Majtan]

Remember, bayer pattern sensors only resolve about 75% of what their raw photo site count indicates.

That greatly depends on how the data is processed...
It is in a way explained here:

http://www.cambridgeincolour.com/tutorials/camera-sensors.htm

I'm not sure how RED exactly processes its RAW data.
Graeme???

:innocent:

 

[Jiri Bakala]

As Jim mentioned, this is not a democracy, so forget polls like this! BTW, they don't work for most so called democracies either, as the governments poll to no end and then do whatever they want anyway

 

[Chris Kenny]

That greatly depends on how the data is processed...

Several tests of Red, including apparently Graeme's, have shown it's 70-80%. This is about as high as it's ever going to get, unless you remove the optical low-pass filter and do extremely aggressive debayering, which will lead to artifacts far worse than a bit of image softness.

So, 3K is just about where you want to be for a 2K or 1080p deliverable, and will give Scarlet a nice edge even over three chip 1080p cameras like the EX1. (Three chip designs don't achieve the full resolution of their sensors either, due to the need for filtering to avoid aliasing, but they come a little closer.)

I'm not entirely sure why there's so much resistance to accepting that oversampling is a really good thing for motion images. In the photo world, people routinely snap 8-12 megapixel images that end up getting displayed as 0.5 megapixel images on web pages, and nobody seems to think anything of it.

 

[David Mullen ASC]

Yes, some of you are forgetting that these are Bayer-filtered cameras, so 4K Bayer is roughly like 3K RGB, and 3K Bayer is like 2K RGB/HD 4:4:4. So I don't think 3K is "overkill" if you want to end up with HD 4:4:4 resolution. Maybe only if all you need is compressed 4:2:2 HD for broadcast...

I think 3K Bayer is about right for a camera that has to deliver high-quality 2K / HD images. 2K Bayer, of course, is being used by some cameras for HD images and it works OK... but you start to push the limits in terms of noise and compression.

But in terms of image quality issues, I'd push for an increase in dynamic range / decrease in compression over all other factors like increased frame rates (but then, that's because my work doesn't involve going over 60 fps very often, and the few times it does, I just rent a specialty camera for that.)

 

[kmikami]

But downscaling also results in less noise and greater dynamic range on a pixel-for-pixel basis, plus you still get a sharper image.

Yep, exactly! I've said it before here and I'm not sure why some people fail to get this. Greater resolution IS greater dynamic range. Plus, you know.. more resolution. :bleh:

I think moving toward higher resolutions is also a cheaper and easier way to get better dynamic range than just working on the chip's sensitivity alone. I mean, obviously you want to be advancing on both fronts. But at a certain point you run into the limits of current technology and issues of heat and size which are interrelated to noise. At that point you might as well squeeze out some extra dynamic range by going to a higher resolution than you need for final delivery.

As a side-effect you also get to reduce any compression artifacts! It's a win win situation.

 

[Peter Majtan]

Greater resolution does not equal greater DR. You will (after and only after down-scaling the image) end up with better noise-to-signal ratio, which can be interpreted as higher DR. But You will never get more details in the highlights and shadows by having more resolution. Period.

Id do get what Chris & David are saying and I fully agree. But put that in the context with what I am proposing in this thread:

http://www.reduser.net/forum/showthread.php?t=12235

And it makes even more sense.

New camera with all the exact internal processing of 3K Scarlet with a new 35mm 3K sensor will nicely fill the void between Scarlet and Red One. It will be the ultimate camera for finishing 2K/HD and could easily squeeze bellow $10K...

It will be nothing like RED ONE, more like Scarlet on steroids with interchangeable lens mounts to compete (and easily win) in the $3-10K range of cameras destined for 2K/HD productions...
It will kill Sony's 9XX series no doubt...

 

[kmikami]

Greater resolution does not equal greater DR. You will (after and only after down-scaling the image) end up with better noise-to-signal ratio, which can be interpreted as higher DR. But You will never get more details in the highlights and shadows by having more resolution. Period.

Dynamic range and signal-to-noise ratio are synonymous. An increase in signal to noise ratio is an increase in dynamic range whether it's brought about by downscaling the image or by other means.

Let's say you stop your camera down to the first point where noise becomes unacceptable. If scaling the image down reduces that noise to an acceptable level then you've essentially gained a stop. You can expose down a stop and save more information in the highlights. This is an increase in usable dynamic range.

 

[Peter Majtan]

Dynamic range and signal-to-noise ratio are synonymous. An increase in signal to noise ratio is an increase in dynamic range whether it's brought about by downscaling the image or by other means.

Let's say you stop your camera down to the first point where noise becomes unacceptable. If scaling the image down reduces that noise to an acceptable level then you've essentially gained a stop. You can expose down a stop and save more information in the highlights. This is an increase in usable dynamic range.

With higher SNR You could get more details in the shadows, but without "actual" higher DR (not just SNR) You will blow Your highlights...

 

[Craig Ryan]

Before we get too caught up in this, I think we need to bear in mind that RED probably has a few new innovative tricks up their sleeve for the new Mysterium X sensors...I'm sure dynamic range is one of them. Plus we still have yet to see Build 16 and all of its image improvement glory.
Aside from that, I am all for a 3k sensor.

 

[Peter Majtan]

PS: Just look at the poll - as of now 75% people agree...

 

[Tyler]

Would having a 3k sensor improve overall resolution at the lower settings, like 1080, when compared to other consumer HD cams?

Or maybe another way to ask the question: would there be any difference in resolution between Scarlet footage shot at 3k and downscaled to 1080p vs. just Scarlet footage shot at 1080p?

 

[kmikami]

With higher SNR You could get more details in the shadows, but without "actual" higher DR (not just SNR) You will blow Your highlights...

No, they're the same thing. If you have lower noise in the shadows you can afford to stop down lower and save your highlights. You simply can't have any useful dynamic range beyond the signal to noise ratio of the system. Therefore an increase in signal to noise is an increase in dynamic range.

 

[Peter Majtan]

Would having a 3k sensor improve overall resolution at the lower settings, like 1080, when compared to other consumer HD cams?

Or maybe another way to ask the question: would there be any difference in resolution between Scarlet footage shot at 3k and downscaled to 1080p vs. just Scarlet footage shot at 1080p?

Absolutely...

 

[Peter Majtan]

No, they're the same thing. If you have lower noise in the shadows you can afford to stop down lower and save your highlights. You simply can't have any useful dynamic range beyond the signal to noise ratio of the system. Therefore an increase in signal to noise is an increase in dynamic range.

I really don't want to argue with You, but this can confuse lot of people.

Yes - You are right that increased SNR results in better DR.
But this is not the same increase in DR as You would get with larger photosites. Let's say camera has a DR of 9. Downscaling actually does not add any DR. It allows You to gently underexpose your shot - which will result in more noise - but then You would "smoothen-out" that noise by downscaling the footage. But when You underexposed your shot - you loose (by same amount) the details in highlights. This would be benefit if Your highlights were blown before. By nature of the sensors the noise only happens mostly in the shadows (due to the lack of light = low SNR).

If You had the same sensor reengineered to DR of 13 You instantly talk different beast. Of course if You then "downscale" Your 13-DR footage You will gain additional SNR improvement. These are two different things.

Where the confusion came from is actually my fault by saying that higher SNR is not higher DR. What I meant to say is the improving DR by rising SNR as a result of DOWNSCALING IMAGE is not the same as improving DR by having larger photosites on the sensor (AKA 3K 35mm chip wersus 4K 35mm chip). My bad - but please do not confuse the two...

 

[Chris Kenny]

Where the confusion came from is actually my fault by saying that higher SNR is not higher DR. What I meant to say is the improving DR by rising SNR as a result of DOWNSCALING IMAGE is not the same as improving DR by having larger photosites on the sensor (AKA 3K 35mm chip wersus 4K 35mm chip). My bad - but please do not confuse the two...

Huh? If you're downscaling such that you eliminate enough noise that you can stop down one stop, you gain a stop in the highlights. In other words, if your dynamic range was previously 9 stops, now it's 10 stops.

Think of it like this. You point a camera at a test chart, and set the exposure so the brightest square is just on the edge of blowing out. Your dynamic range is entirely determined by how many distinct squares you can count before noise swamps the image sufficiently that you can't tell where one ends an the next begins. If there's less noise (it doesn't matter how you get less noise), you can count more squares. Your dynamic range is greater.

 

[kmikami]

What I meant to say is the improving DR by rising SNR as a result of DOWNSCALING IMAGE is not the same as improving DR by having larger photosites on the sensor (AKA 3K 35mm chip wersus 4K 35mm chip).

They are two different techniques to achieve the same ends. It's technically true to say that they're not "the same" but the extra dynamic range you get from downsampling is just as good as the extra dynamic range you get from a better chip. There's not good and bad dynamic range.

I think you're confusing dynamic range as limited by the noise floor with dynamic range as limited by the bit depth. The problem is, as far as I know, with the bit depths we're dealing with, the noise floor is always the limiting factor. In which case improving the signal to noise ratio is the way toward improving dynamic range.

If we somehow got into a theoretical situation where there was a totally noiseless system and the bit depth of the system was the only factor limiting the dynamic range then perhaps downsampling wouldn't do any good.

 

[Peter Majtan]

Huh? If you're downscaling such that you eliminate enough noise that you can stop down one stop, you gain a stop in the highlights. In other words, if your dynamic range was previously 9 stops, now it's 10 stops.

Think of it like this. You point a camera at a test chart, and set the exposure so the brightest square is just on the edge of blowing out. Your dynamic range is entirely determined by how many distinct squares you can count before noise swamps the image sufficiently that you can't tell where one ends an the next begins. If there's less noise (it doesn't matter how you get less noise), you can count more squares. Your dynamic range is greater.

Yes, it is.
But this is NOT the same thing as having a sensor with physically higher DR.
Period.
The two methods are complementary - but certainly NOT THE SAME as You trying to say. This is very confusing to people who look here for knowledge...

You have a sensor A and sensor B, both 3K, but A is 2/3" and B is 35mm.

B is going to have higher DR (with less noise) then A.
You can improve both DRs by downscaling the two images to 2K, but the image from sensor B will still have higher DR.

I can't be any more clearer that this (for lack of my English skills or whatever reason), sorry...

 

[Radoslav Karapetkov]

WTF is this inside-outside USA stuff?