2K scaling...

[Mark Pugh]

can 3k be offered at a maximum frame rate somewhere between 60 and 100?
And recording times greater than 4k?

 

[Hans von Sonntag]

Seems to me that 2K scaled overcranked is around the corner. Just, it's not 2K but 4K... Not too bad.

BTW, recordings times are not an issue, by then...

Hans

 

[PaulClements]

Well if we've been asking for the wrong thing what about the following

• Variable Framerate (With a minimum of 60fps as the maximum)
• Variable X to Y pixel selection on the sensor (i.e. you can map the sweet spots of lenses such as anamorphic, fourthirds, S16mm/16mm etc).
• Variable RedCode Compression ratio
• Variable Scaling
  

I'd prefer to have all the above options completely open and changable in camera, with the maximum fps increasing or decreasing dependent on the amount of pixels used, the level of compression selected or the amount of scaling involved. Do away with the entire this format or that format and let the users define their own. For example 1080p could be selected as 3840 pixels (X) x 2160 pixels (Y) scaled by 4, the camera works out how much processing this will take and modifies the framerate accordingly - Don't have a high enough framerate? Then drop the compression down and raise it until it fits the job - then review whether the footage looks ok or not.

Being able to define presets so that one could map out individual lenses or regularly used formats would be essential.

I know this isn't something that would be very easy to do, but fact is it ought to be possible and would make the camera very versatile. I wouldn't look at scaling and altering ratios in camera as a bad thing as such, look at it as offering greater scope and modularity which was always at the heart of the cameras conception after all.

Paul

 

[redrum]

- 2K scaled requires 4 times less bandwidth, so you can record 4 times more footage on the same media.

Are you sure about that? I wouldn't be.

We're talking about compressed data, not raw. The compression may not scale linearly. For example, I grabbed a 4K still and saved it as a 4K JPG (which took 4.8 megabytes) and as a 2K JPG. You'd think the 2K JPG should take up 1.2 megabytes, right? Bzzzt. It was 1.7 megabytes. Even though the source 2K image represented exactly 1/4 the data of the source 4K image, the resulting compressed size of the 4K JPG was not 4 times the size of the 2K JPG; it wasn't even 3x. It was more like 2.8x. And this was with old nasty JPG compression, not fancy new wavelet compression. It may be that the scaled wavelet-compressed image wouldn't result in anything approaching 1/4 the size, it might be 1/3 or it might be half. In any case it should be less than the size of 4K, yes, but it won't be 1/4.

I don't know exactly why, but you're not gonna get 4x as much data, you'd probably be looking at maybe 2x as much with the more-efficient wavelet compression.

All irrelevant anyway as Jim already said it won't happen, no scaled image in-camera.

- You may be able to use slower recording media (CF, SSD, etc...)

Valid point. But again, not 4x different, more like 2x. So you may not require 40MB/sec flash media, but you sure aren't going to be able to get away with 10MB/sec! It'll be more like needing 20MB/sec. A difference, sure, but not as much as you'd hope.

- Processing times in Redcine should be much shorter from a 2K source, than from a 4K source.

Totally valid.

Or do you mean you just can't do it due to hardware limitations ?

That's the way I'm reading it.

 

[redrum]

Hey, just a thought -- what if they just added variable compression/bandwidth levels to redcode. If you chose chunky low compression at half the data rate you'd probably get what you want -- 2K worth of data, longer record times, and the ability to use cheaper media. It'd still be recorded in a 4k frame size, but the compression would be dialed back and all you'd likely lose would be the highest-frequency detail -- but you're going to lose that when you scale down to 1/4 the size anyway, so -- would that work?

 

[PaulClements]

Hey, just a thought -- what if they just added variable compression/bandwidth levels to redcode. If you chose chunky low compression at half the data rate you'd probably get what you want -- 2K worth of data, longer record times, and the ability to use cheaper media. It'd still be recorded in a 4k frame size, but the compression would be dialed back and all you'd likely lose would be the highest-frequency detail -- but you're going to lose that when you scale down to 1/4 the size anyway, so -- would that work?

One of the reasona for scaling in the first place is to lose any compression artifacts that occur at 4K. If you are adding in more at 4K then the scaled version would be less valuable.It also means you still have to handle 4K footage in post.

That aside having access to as many variables as possible (See my last post ) is never a bad thing, it all adds to the mastery of the tool.

Paul

 

[Obin Olson]

2k is a good option for people that only need 1080p or less as the end result. I would shot lots and lots of stuff 2k as a professional just because I don't need that extra res, and I sure don't need the extra render times!

 

[Nils Ruinet]

Are you sure about that? I wouldn't be.

We're talking about compressed data, not raw. The compression may not scale linearly. For example, I grabbed a 4K still and saved it as a 4K JPG (which took 4.8 megabytes) and as a 2K JPG. You'd think the 2K JPG should take up 1.2 megabytes, right? Bzzzt. It was 1.7 megabytes. Even though the source 2K image represented exactly 1/4 the data of the source 4K image, the resulting compressed size of the 4K JPG was not 4 times the size of the 2K JPG; it wasn't even 3x. It was more like 2.8x. And this was with old nasty JPG compression, not fancy new wavelet compression. It may be that the scaled wavelet-compressed image wouldn't result in anything approaching 1/4 the size, it might be 1/3 or it might be half. In any case it should be less than the size of 4K, yes, but it won't be 1/4.

I don't know exactly why, but you're not gonna get 4x as much data, you'd probably be looking at maybe 2x as much with the more-efficient wavelet compression.

That's what I would have thought too, but Jim said a few times that the 2K bitrate was about 4 times less than the 4K bitrate :

From the "I was wrrr...wrrrrong." thread :

.....

4. The record time to 8GB CF is over 20 minutes.

.....
Jim

From the "720P and 1080P... losers" thread :

....Remember that shooting 2K REDCODE RAW is only about 7MB/sec.

So even if the datarates are a bit higher now, this is still about 1/4th of 4K.

Being able to choose a higher compression, why not, but my wild guess is that you would have a better quality and less artifacts using 2K scaled... And you would still have to handle 4K files in Redcine...

By the way, Jim, did you change your mind since you wrote this ?

OK. I have to eat crow. I admit it. I won't just shoot 4K.

Shooting 2K windowed is a completely valid option.

1. Even windowed from 4K, the 2K sensor portion is larger than 2/3", better DOF characteristics than an F23, HDCAM or Viper.

2. Having 72fps (for now) at this frame size is better than HDCAM-SR 4:4:4, which now shoots 60fps at 1080P.

3. You can use all your 35mm PL mount lenses, except at the wide end due to the focal length factor. You'll need a wide S16 lens or two to cover all the bases.

4. The record time to 8GB CF is over 20 minutes.

5. Pushing the data through the pipeline is much easier.

And best of all, the image looks great.

I still will shoot most of my stuff 4K, but I won't hesitate to shoot 2K windowed if necessary.

Jim

2K scaled being suposedly even better than windowed 2K, why would you suddenly want to drop it ?

Nils.

 

[redrum]

Interesting. Seems wavelet is more directly scalable per size of frame?

2K windowed is easy. 2K scaled is not, especially if the existing DSP doesn't include such functionality (don't know that).

Unless the sensor itself supported pixel binning. Lots of sensors do. A simple 2x2 pixel bin would get you 2K resolution and also make the Red effectively 4x more sensitive to light, and if I'm not mistaken you could read the sensor twice as fast too (so you'd have 2K resolution, 120fps speed, and ISO 1280).

Jim? Graeme? Does the Mysterium support a 2x2 pixel bin? Could this be used for onboard 2K? 'course, the bayer pattern probably ruins the potential for binning; it may only work properly for monochrome sensors. Or you may end up with 1280 ISO 2k @ 120fps, but only in black and white. Which would still be pretty cool.

 

[Axel Mertes]

Interesting. Seems wavelet is more directly scalable per size of frame?

2K windowed is easy. 2K scaled is not, especially if the existing DSP doesn't include such functionality (don't know that).

Unless the sensor itself supported pixel binning. Lots of sensors do. A simple 2x2 pixel bin would get you 2K resolution and also make the Red effectively 4x more sensitive to light, and if I'm not mistaken you could read the sensor twice as fast too (so you'd have 2K resolution, 120fps speed, and ISO 1280).

Jim? Graeme? Does the Mysterium support a 2x2 pixel bin? Could this be used for onboard 2K? 'course, the bayer pattern probably ruins the potential for binning; it may only work properly for monochrome sensors. Or you may end up with 1280 ISO 2k @ 120fps, but only in black and white. Which would still be pretty cool.

I think there is no active binning on the sensor, because - as you already mentioned - it makes few sense in color Bayer pattern. There are sensors out there which can do some kind of binning per Bayer quadrant, that could have helped, but I doubt its in, because then they would for sure use it already.

For these reasons I am sure the maximum frame rate is 60Hz full sensor, and half the lines = double speed = 120Hz windowed. I wonder if the window could be adjusted "at will" as many industrial cams do. So you can do even higher frame rates using lower areas etc. Rough estimate: A true (tele style windowed) PAL would yield about 250Hz - not that bad. Probably nothing for which Jim is looking, BUT we users...

Anyhow, if a chip is Bayer, you can't use it monochrome...

What I really wonder is Jim's this morning note why we ask for scaled 2K from full sensor instead of full 4k REDCODE at full 60Hz. It amazes me that the fully binary downscale from 4K to 2K should be slower than compressing 4 times as much wavelet pixels. Honestly I really doubt that, computing-wise. But I would be happy to learn different...

Maybe someone from RED chimes in and shed some light on this.

I would still want to have the 2K scaled option, if at any possible.

Cheers,
Axel

 

[Fergus Meiklejohn]

I suppose we could film docos in 2K windowed with S16 lenses or with the canon 10-22, 17-55 combo; covering the same range as the Cooke 10.4-52.
The 16mm field of view characteristics would also make focussing easier..

I would want to be able to switch between 2K windowed and 4K very quickly (without rebooting etc): so the idea is that I would film GV's 4K, for the extra quality and so I can crop them loads in post to get a much tighter shot for example, and I might film a rambling hand-held interview 2K windowed..

Then downconvert the whole lot to 1080P (eg. Prores), for editing and finishing.

And maybe in two or three years we'll have massive Compact Flash cards and such fast laptops that this whole debate will become redundant..?

 

[Dave Neathery]

Personally, 4K at 60fps is exactly what I have been wanting. That is perfect for most of what I plan to do.
Dave

 

[Álex Montoya]

4K at 60fps is perfect

 

[Matt Gottshalk]

2k is a good option for people that only need 1080p or less as the end result. I would shot lots and lots of stuff 2k as a professional just because I don't need that extra res, and I sure don't need the extra render times!

Exactly what I was thinking!

 

[Blair S. Paulsen]

4K fan

4K fan

For the sake of this post I am stipulating that full quality 2K scaled is not a viable in camera "on the fly" operation.

I have become convinced from viewing actual footage from the RedOne that the implementation of Bayer pattern CMOS tech with OLPF out front and algorithms that control moiré is roughly 2.8K in perceptible resolution at nominal 4K and around 1.4K at nominal 2K. Because of this I would not consider 2K windowed to be appropriate for a high quality 1080P finish - of course in many cases the actual distribution path would make such resolution distinctions impossible to see but that is another discussion.

My real point is that the RedOne really shines when you acquire at nominal 4K even if you end up down rezzing to 2K or 1080 for finishing due to the huge advantages of oversampling. The are a number of reasons why the images from the RedOne look so "filmic"; cine lenses, 35mm DoF, RAW color space and the subtle gradations of density and tone made possible by oversampling.

Bottom line, even if 4K seems like overkill I contend that it is worth the extra overhead - IMHO you, and your clients, will see and "feel" the difference.

 

[Jeff Kilgroe]

IMO, the only reason to want 2K scaled is for the ability to have the abilities of the full 35mm FOV / DOF, etc.. But allowing for on-board recording at rates up to 60Hz.

However, I'm not so sure that it will be possible. The reason we don't have 4K @ 60Hz on-board is because the internal processors of the RED One can't cope with that amount of data. If that's the case, in terms of processing to REDCODE, how is it going to deal with that same amount of data, sample to 2K, and then encode to REDCODE (although at a lower res and data rate)? In my mind, that doesn't add up.

And I agree with Blair. This camera is meant to shoot 4K. 2K in a pinch if you need the overcrank ability or if you want to use existing broadcast or 16mm glass and HD quality will do.

Record times shouldn't be a real concern here. Real productions are being made just fine with 8GB CF cards when shooting 4K. Most of the concerns over longer record times won't be solved by shotting 2K to those same CF cards and it will be at a huge compromise in other areas. RED DRIVES and potentially other solutions are just around the corner.

 

[Gavin Greenwalt]

I'm still not convinced (but have seen no evidence one way or another) that the resolution hit from oflow retiming is worse than shooting 2k scaled bayer. (2k scaled RGB is pointless since it's practically the same datarate and more computationally intensive in-camera).

With lots of thin smoke you might have problems with stuttering but for the most part just getting the slo-mo look probably would be better in post and losing 30% your resolution than the full 50%+ from resampling, blurring and rebayering.

 

[David Mullen ASC]

Let's face it, 2K scaling is a compromise, to deal with limitations in processing speed and data storage. Ideally, the camera would just stay in 4K mode all the time, and everything else in the chain would improve to handle that. Getting a good image with a Bayer-filtered sensor and a OLPF is hard enough without throwing in real-time scaling in RAW down to 2K. Ideally, you'd be able to increase storage to handle more 4K footage, or processing speed to handle 4K at higher frame rates, and just skip the idea of scaling 4K to 2K in RAW mode in camera.

 

[Jannard]

Let's face it, 2K scaling is a compromise, to deal with limitations in processing speed and data storage. Ideally, the camera would just stay in 4K mode all the time, and everything else in the chain would improve to handle that. Getting a good image with a Bayer-filtered sensor and a OLPF is hard enough without throwing in real-time scaling in RAW down to 2K. Ideally, you'd be able to increase storage to handle more 4K footage, or processing speed to handle 4K at higher frame rates, and just skip the idea of scaling 4K to 2K in RAW mode in camera.

Amen...

Jim

 

[Greg M]

personally if I could shoot 4K at any frame rate from 1-60, I wouldn't ask for anything else.

well...actually I would, but I'm greedy.