Is 1080p Useful? Science Provides an Answer. (Hard Data Within.)

[Gavin Greenwalt]

Process:

In order to test the rescale render performance I used Fusion on a theoretical top of the line system (divided all results by 10 to more accurately predict what it would be like on a new workstation, not a POS 4 year old "check your email" desktop.)

Process was to import an uncompressed 4k image sequence. Crop it to either 4k, 2k and then rescale (using a bi-linear rescale) to 1080p. The sequence would be rendered and the average render time per frame would be extracted. In order to avoid outliers 5 'samples' per size were taken in sets of 2. In other words I would do two 4k, two 2k and two control renders before starting over. Once complete the Maximum and Minimum result were discarded and the remaining 3 median results were then averaged.

In order to remove disk access and decode performance from the test I also created a control value by simply opening cropping and closing the file.

All benchmarks are the difference between the actual number and the control number (and then divided by 10 since my home computer really really sucks. The number 10 was determined by looking on Toms Hardware's CPU charts and seeing how mine compared to an Intel Q6600 in 3D Studio Max Render benchmarks).

As a reference mark I also compressed the same clip from the control setup to a 1080p Windows Media Video file (Mpeg 4 v3) in order to simulate saving out to DVCPro or some other deliverable codec.

Data:

4k scaled to 1080p: .07 Seconds Per Frame (14.28 frames per second)
2k scaled to 1080p: .0197 Seconds Per Frame (50.76 frames per second)

1080p to Mpeg 4: .0837 Seconds per Frame (11.94 frames per second)

Conclusions:

The time difference between 1080p RGB and 2k RGB resized to 1080p is negligible. Especially in comparison to to the render times of MPEG 4 which in this case took longer than even scaling from 4k RGB to 1080 RGB.

Notes:

This was actually my second round of testing. I finished and then had to slap myself in the head. I used a "resize" operator which is an arbitrary node and not optimized for a linear scale. Results are from a "rescale" node.

However doing it twice and with two different algorithms did offer some insights. In both tests the proportions between 2k and 4k were almost identical. I performed the MPEG - 4 test the second time as a difference between the 4k scale and the 4k scale saved out to MPEG4. The difference between both approaches was less than 4% which adds further credence to that result.

This test doesn't take De-Bayer or Wavelet decode time into account. For more information on Wavelet decode performance I will refer the reader to my article in the latest showreel.

Also it should be noted that REDCine is GPU accellerated but DVCPro, Apple ProRes and DNxHD compression are not. So you should expect even better performance than this with a fast video card but your deliverable encode times will remain static.

My personal opinion is that 1080p is not a useful format. I leave it to you to decide based on this data if you agree.

[Brook Willard]

These are pretty interesting data... interesting enough to make me do my own test. If yours is the high end, mine's the low.

Using my Quad G5, I created three sets of frames: 1080p, 2K and 4K. I compressed the frames to approximate REDCODE's data rate [~24MB/s for 4K, etc.]. I made 300-frame sequences of all three and exported from Quicktime Pro to 1080p using several different HD compressors. The results that follow are from XDCAM HD.

1080p to 1080p XDCAM HD: .1443 seconds per frame [6.93FPS]
2K to 1080p XDCAM HD: .1596 seconds per frame [6.26FPS]
4K to 1080p XDCAM HD: .3926 seconds per frame [2.54FPS]

Obviously these results are not necessarily indicative of what REDCODE's export times will be... not in the slightest. It does show a little something about the difference between a scale/export and a 1:1 export, though... and scaling from 2K to 1080p is not exactly a mind-blowing task for the average computer.

[Alexander Nikishin]

Finally some tests, great to know. :greedy:

[Jeff Kilgroe]

I started to do some testing of my own yesterday... And then didn't. I really don't know what we can expect from REDCINE and REDCODE at this point. And I don't have any encoding software that is GPU enhanced or even multithreaded to the point of efficiently using more than 2 CPUs. So at this point, it's all speculation. But I think these preliminary test numbers with other software could give us a glimpse. If nothing else, it gives us an idea of what to expect when we transcode RED footage to something else... In most cases, I still think the limiting factor is going to be whatver codec software we are transferring into.

[David Battistella]

Process:

My personal opinion is that 1080p is not a useful format. I leave it to you to decide based on this data if you agree.


Gavin,

Thanks for doing these tests and posting the results. I know it's not REDCINE, but in some respects data crunching is Data crunching, so it does offer some very useful insights.

I bet you are closer than you think with these results and times, and if you are, then 1080P as an "in camera" recording format makes less and less sense.

Jim mentioned in an earlier post that not long ago, people were asking the camera to record SD. HD might not be "recorded" but you can always cut out or scale 4K or 2K to anything and that is incredible flexibility.

I was illustrating this to a friend in a bar the other night just by using a paper napkin and folding several times (each fold representing the downsizing of an image and representing the various available formats).

I'd take an 8x10 negative over 110film any day and I think that philosophy applies here too.

David

[Jack Wester]

Thanks for sharing these results Gavin.

As downsampling a bitmap is very simple stuff I'd assume that it has been optimized for speed in Fusion. But if downsizing is a part of everyday workflow, and its code is not optimized it makes a very good candidate for threading. This means that it should scale almost linear with the number of CPU cores (one thread working with one frame at a time and running as many threads as there are CPU cores).

Did you test if the code was multithreading. A simple way to test this would be to run two instances of the resampling software on a multicore computer to verify that the frames per second per instance is less than half. If they are, the code was multithreading to begin with. Otherwise you can expect better results as most will have modern CPUs or multiple CPUs.

As it should scale almost linearly, the results are more usefull if you specified how many cores was present.

[Gavin Greenwalt]

I can guarantee it wasn't multithreading. :D

[Jack Wester]

I can guarantee it wasn't multithreading. :D

Are we talking about resampling in post? Writing code that resamples using multiple threads is really easy, it'll be the RedCode compressing that consumes time.

I've done rescaling code more times than I can count. The two major hussles of multithreading are

1. Concurrency issues in sharing and updating (locking) data between threads
2. Linear processing. Results that depends on other results.

Neither of these problems are present when resampling. Especially not with multiple frames as the case with film as apposed to still images (even one still image is easy, just divide it to as many sections as there are cores with some overlap). Writing that code is _really_ easy and would deliver its output without to the encoding subroutines without the encoder code having to know about any prior resampling. 1080 should simply not be needed in the camera or the resampling code should simply we put into the camera as a prepost operation.

[Peter McCully]

Process:


Process was to import an uncompressed 4k image sequence. Crop it to either 4k, 2k and then rescale (using a bi-linear rescale) to 1080p. The sequence would be rendered and the average render time per frame would be extracted. In order to avoid outliers 5 'samples' per size were taken in sets of 2. In other words I would do two 4k, two 2k and two control renders before starting over. Once complete the Maximum and Minimum result were discarded and the remaining 3 median results were then averaged.



I'm a Fusioneer also. What algorithms were you using in the resize node, and what gave the best result?
Also, How in Fusion did you achieve the sample sets and median results?

[Jack Wester]

The major amount of time spent in the process when going from 2K to 1080 or 4K to 1080 would be decoding the compressed RedCode and encoding the 7 MB/Sec RedCode. Not the resampling. So is this resampling benchmarks really relevant? Or am I getting tired?

[Stephen Webb]

It's an interesting test, and helps to answer a few questions, but Jack's comment that the major processing task is uncompressing the Redcode & then recompressing to whatever format is right I think. In that case though 2K Redcode RAW is likely to hold an advantage over 1080p Redcode RGB - how much that is then offset by the need to debayer & process the RAW is uncertain.

[Rocco Schult]

1080p to 1080p XDCAM HD: .1443 seconds per frame [6.93FPS]
2K to 1080p XDCAM HD: .1596 seconds per frame [6.26FPS]
4K to 1080p XDCAM HD: .3926 seconds per frame [2.54FPS]

... and scaling from 2K to 1080p is not exactly a mind-blowing task for the average computer.

Thanks for the test and bringing the numbers into the game Brooke.

What I'd like to consider:
- time of scaling (Quicktime does the worst scaling you can imagine, its about bilinear or something and thus not acceptable as scaler for using the pictures other than preview imho) It might not be a 'mind-blowing' task, but it is a calculation. Hence:
- the quality. If the engine you decide to go for includes bicubic scaling, what I propose it should do and the question of oversampling and/or filtering (e.g. lanczos) not spoken here, then I am sure it does affect processing time more significant.
And consider what you are resizing. You try to munch 100 and something pixels in a 1920 picture. Its not a 2:1 or 4:1. Its a 1.066:1 or so. You'd better do that good or it will look pretty odd.

I have evaluated scaling time a while ago for a movie, scale-up from SD to HD and we wanted to have the best scale we could think of, being capable of and doing it asap. We ended up with double- and tripleshifts, because we didn't want to go for less then the best we could achieve. There was no debayer etc. - just a "good" scale. Even if opposite direction, I think a good scale takes time (even if CPUs got faster in the meantime, hell, they always do..)

And generally, taking the files without any scaling has to be faster than with, right ? I'd appreciate at least 1 mode like the much discussed 1080p, which just is and needs as less as possible processing. I don't talk of 720, that sucks.-)

But in the very end it comes down to the same question over and over again:
How long does post-processing/resizing/RedCine take.
If its a nobrainer, its a nobrainer. We had that several times here now.
I think we hear nothing for good reason. It will giv us pain until the machines get better and faster. Thank God they do.
On the other hand, time was working for everybody ordering the RED, why shouldn't it be good for the developers too.:wacko:

[Gavin Greenwalt]

I actually never looked at the renders. In order to keep things limited to "rescale time" my flow was:

Loader: 4k uncompressed Quicktime
Crop: 2k/4k resolution
Scale: 1920/2048 or 1920/4096 Bilinear
Crop: 4px by 4px
Saver: Tiffs or MPEG 4

Originally I was using a Resize instead of a scale. It was about 4x slower.

Render times were determined by using the results of the render. It tells you the average time per frame in the status bar. I would then record it to a notepad file. Render. Flush Cache. Record time. Render flush cache... record time...

Scale Node performance was determined by rendering once with the node on and one with the node on pass through. Find the difference of the averages (ignoring the max/min value from each sample set).

----

What I meant by "I know it wasn't multithreading" was I was using an Athlon XP 2100+

----

The objective of the test was to determine to what extent scaling will contribute to render time. It seems worst case scenario it'll probably be less than 30% and best case scenario negligible.

However based on the "Candy Bar Challenge" it seems 2k RGB -> 1080 RGB scale may not be a viable option for quality reasons. In which case I don't need to perform any tests to say that a 2k -> 1080 crop is instantaneous.

----

Just did a Bicubic vs. BiLinear benchmark.

Bicubic took about 30% longer. (1.3 times bilinear's time)


----

Finally this is an oldy moldy but it's kind of the definitive guide to scaling:
http://www.fxguide.com/article315.html

[Graeme Nattress]

I think the best way to master to 1080p is to master to 2k and centre-extract? Why? Well, with HD monitors there is no overscan. Unless you're very careful, practically any image processing operation may fail to be totally consistant at an edge border of the screen. By mastering to 2k and cutting out a 1080p you can ensure that there are never going to be any nasties on the very edge pixels that can be seen with a native 1:1 display. Thoughts?

[David Battistella]

I think the best way to master to 1080p is to master to 2k and centre-extract? Thoughts?

This makes a lot of sense and this seems to be the point of shooting the larger frame sizes. Center extraction may take some time out of the post process too, since you'd eliminate any scaling work. You also aren't softening the image in any way.

I would suggest that if this were to become the option, RED includes a foolproof EFV and "in viewfinder" monitoring solution which includes accurate guides that would make the shooting process even more seamless.

Could this go into a metadata flag as well?

Could the codec be asked to "playback" only a center cut?

If you knew exactly what you were center cutting you could shoot 2K with HD 1080P center cut guides ON and feel very confident about the result.

This might be obvious but it seems like a smart, simple way to go. I would guess that you suggest this because it's one way of reducing, compressing or doubling the time the footage needs to go through REDCINE?

I am sure that many will say "but I want to use the full sensor", but since most of the 1080P discussion seems to be centered around "day of" speed and "deliver ability", then can you have you cake and the icing too? :) I'd venture a guess that the center cut image from the Mysterium Sensor looks better than any system that currently shoots and records 1080P.

David

[Phil Bates]

I think the best way to master to 1080p is to master to 2k and centre-extract? Why? Well, with HD monitors there is no overscan. Unless you're very careful, practically any image processing operation may fail to be totally consistant at an edge border of the screen. By mastering to 2k and cutting out a 1080p you can ensure that there are never going to be any nasties on the very edge pixels that can be seen with a native 1:1 display. Thoughts?

Exactly right. Also, it leaves room on the edges for stabelization adjustments.

Phil
www.artbeats.com

[Sean Michael Johnston]

Cropping from 2K to 1080p seems the to be the best solution. You can record raw and have a little room for motion stabilization. Cropping takes alot less time than resizing and no softening occurs.
http://www.reduser.net/forum/uploaded/709_1186328521.gif

[Michael Morlan]

I think the best way to master to 1080p is to master to 2k and centre-extract? Why? Well, with HD monitors there is no overscan. Unless you're very careful, practically any image processing operation may fail to be totally consistant at an edge border of the screen. By mastering to 2k and cutting out a 1080p you can ensure that there are never going to be any nasties on the very edge pixels that can be seen with a native 1:1 display. Thoughts?

Graham, if there is no 1080p on the camera, then I agree absolutely. Simply crop rather than scale.

One thing that hasn't been discussed in all this scaling business is loss of resolution due to the Kell Factor during the scaling (not to mention displaying) of 4K->2K->1080p->720p->480p images. All that inter-pixel interpolation, unless you're scaling at some integer multiple, causes additional loss of detail after the scale.

M

[Simon Blackledge]

another hand raised for "crop"


s

[Stephen Webb]

Graeme,

Is it possible to have the crop automated? If there is a 1080 framing guideline in the viewfinder (& other monitors) and, as others have suggested, a metadata tag that simply tells FCP/Redcine to ignore the edge pixels this could be a real winner.

[Storyline]

Graham,
What is the method for cropping? Is this something that RedCode can tell FCP and PremierePro and do on the fly? Will it then need to be rendered? Is it a RedCine process pass? How fast would center extraction be?

[Nils Ruinet]

I think the best way to master to 1080p is to master to 2k and centre-extract? Why? Well, with HD monitors there is no overscan. Unless you're very careful, practically any image processing operation may fail to be totally consistant at an edge border of the screen. By mastering to 2k and cutting out a 1080p you can ensure that there are never going to be any nasties on the very edge pixels that can be seen with a native 1:1 display. Thoughts?

Yep, sounds fine...
With all these latest news I'm really starting to think that losing 720P / 1080P in camera wouldn't be such an issue...


Is it possible to have the crop automated? If there is a 1080 framing guideline in the viewfinder (& other monitors) and, as others have suggested, a metadata tag that simply tells FCP/Redcine to ignore the edge pixels this could be a real winner.
I second that too.

And yes, would the quicktime codec be able to handle that on the fly for preview and editing in FCP ?

[Graeme Nattress]

We can probably tell the codec to crop based upon metadata. I don't think the codec will do that off the bat, but again, it's a good thing to put "in the list".

Kell factor durin scaling? Kell effect is how images displayed on a monitor appear less sharp than they really are due to the reconstruction filter of the beam spot as it scans the CRT. I don't see what that's got to do with scaling?

Graeme

[Adrian T.]

Graeme,

the coolest thing would be to have another proxy for 1080P which crops from 2K to 1080P (the NLE knows nothing about the extra pixels, it's a simple 1080P video stream). So when time is short, it's possible to go directly into a NLE using this proxy, without a conversion pass in REDCINE.

Even cooler would be another proxy for a 2K extract plus 1080P crop from a 4K file.

What do you think?

[Joel Kaye]

I think the best way to master to 1080p is to master to 2k and centre-extract? Thoughts?

Makes a lot of sense to me.

I think if there's any way we can import our own viewing guides that would be really helpful. Let us load up a .png an overlay it. We could build custom eyeline and composition guides as well as guides for VFX shots. Heck, you could build guides from previz/storyboard frames and toggle them on/off.

That's just gotta be doable.

[GlennChan]

I think the crop method makes more sense. It gives you more flexibility (e.g. true 2K master) but should practically be the same performance and data rates no?

[Michael Morlan]

We can probably tell the codec to crop based upon metadata. I don't think the codec will do that off the bat, but again, it's a good thing to put "in the list".

Kell factor durin scaling? Kell effect is how images displayed on a monitor appear less sharp than they really are due to the reconstruction filter of the beam spot as it scans the CRT. I don't see what that's got to do with scaling?

Graeme

Thanks Graeme,

Yep, I knew "Kell" wasn't the right term for scaling, just displaying, but the effects are the same as I understand it. Is there a term for resolution lost to scaling?

M

[guinwhite]

perhaps a question that has been addressed prior, but does RedCine come with the Red One, or is it purchased separately?

[Steven Parker]

perhaps a question that has been addressed prior, but does RedCine come with the Red One, or is it purchased separately?

Comes with the Red One. Whether it's available separately is still TBD.

If the numbers that opened this thread are a fair estimates, then losing 1080p in-camera is not such a big deal... 2K crop makes a lot of sense...

[Graeme Nattress]

Well, scaling implies resolution loss as you're going from, say, 4096 to 2048 pixels. You've just lost half your resolution - or have you? It all depends on how the downscale is done and I tend to find that a good downscale can pack in more information into those pixels as now you're oversampling. A good downscale should be per-pixel sharper, but with less pixels.

Graeme

[Joel Kaye]

It all depends on how the downscale is done and I tend to find that a good downscale can pack in more information into those pixels as now you're oversampling. A good downscale should be per-pixel sharper, but with less pixels.


That's a good point. Whatever noise there is at 4K could be reduced when downscaling. Reminds me of the old comments that RED will be a tremendous 2K camera because it shoots 4K.

It'll be fun to see all the tests everyone does to get everything figured out.

[Vincent Rice]

Downscale 4K to 2K. Post process 2K. Crop 2K to 1080. Perfect. I can stop thinking about this now.:mellow:

[Blair S. Paulsen]

I think the best way to master to 1080p is to master to 2k and centre-extract? Why? Well, with HD monitors there is no overscan. Unless you're very careful, practically any image processing operation may fail to be totally consistant at an edge border of the screen. By mastering to 2k and cutting out a 1080p you can ensure that there are never going to be any nasties on the very edge pixels that can be seen with a native 1:1 display. Thoughts?

Having had several front projection systems fed by a wide variety of sources over the years I have seen some very interesting errata on edge pixels. Those of us who were creating web versions of video content in the early days often found some elements in the last few rows that the producers assumed would never be seen. Now don't get me started on 480 vs 486, SD is so yesterday :nuke:

Bottom line - I love this approach, in part because I expect image stabilization at the post production stage to rapidly improve. Apple's "smoothcam" tech is just the beginning. With better algorithms, direct GPU topologies and those extra pixels we could have amazing motion tracking results. What a great stand out feature for the RedOne over straight 1080 cameras that would have to scale, with the attendant artifacting, to achieve the same thing.

Now, if I've sparked your imagination, how about the equivalent of ease-in ease-out with editable curves, selectable lag... Having shot a lot of handheld I really like the idea that most of my footage would have just the "look" the director asked for but the little bobble stepping off the curb or over the cable or bumping against the grip, etc could be seamlessly fixed, nice!

Steadicams are great but if I can shoot handheld, pull the selects and circle takes at the end of the day, process them overnight on commodity hardware and have rock solid footage in the editors hands the next day I will be one happy camper. (Now maybe Tiffen can get their focus back on making filters :sarcasm: )

[Gavin Greenwalt]

Apple's Smoothcam is already pretty significantly outdated.

Mind you the new solutions are more expensive and slow, but it's still outdated none-the-less.

[Häakon]

Downscale 4K to 2K. Post process 2K. Crop 2K to 1080. Perfect. I can stop thinking about this now.:mellow:
If you crop your image, you are losing FOV and wasting sensor resolution. Why not just downscale to 1080 and skip a step?

[Brook Willard]

If you crop your image, you are losing FOV and wasting sensor resolution. Why not just downscale to 1080 and skip a step?

I think he's referring to the 4K-2K scale happening in camera.

[Häakon]

I think he's referring to the 4K-2K scale happening in camera.
I am too. :-)

[Alexander Nikishin]

Blair, I don't think hand held will ever match what can be done with a Steadicam.

First off, Steadicam's can do low mode, try keeping that stable hand held. Steadicams can also track while raising or lowering from lo - high mode, that's just not possible if you're going hand held.

My problem with stabilization programs is that you lose part of the image if you're correcting jittery camera work, that's unacceptable for most.

I really value a good Steadicam Op., they are worth their weight in gold and are often paid in gold on set.

[Brook Willard]

I figure the concern has to do with processing time and potential softening when making such a minor resolution change with a badass [read: nicer looking than the Quicktime] scale.

[Gavin Greenwalt]

My problem with image stabilization is motion blur (sort of correctable using oflow) and parallax (uncorrectable). Parallax is the real killer. Unless we develop some sort of incredible image based 3d scene reconstruction techniques steadicams are here to stay.

[Häakon]

I figure the concern has to do with processing time and potential softening when making such a minor resolution change with a badass scale.
I agree 100%; that's precisely why a 1080p scale is better. You don't want to end up either softening your image or losing part of it to a crop (as Alex just pointed out, "most will find that unacceptable"), and since no one is going to be finishing in 2K, it's inevitable that one of those two things has to happen with a 2K scaled solution. It's the point I've been trying to make for a week now. The only drawback may possibly be processing time, but quality trumps time in my book any day.

[Jeff Kilgroe]

I wouldn't say no one will finish in 2K. Anyone doing 2K DI finishes or output for 2K DCI / projection will finish in 2K. For any of my shorts that are destined (hopefully) for that route, I will definitely finish 2K and only do a scale to 1080p when mastering for BluRay, HD-DVD or broadcast. I actually intend to most of my finishing in 2K until software catches up and 4K finishing becomes more practical. :bleh:

I do agree with the need for a 1080p direct scaled option though. It would probably serve many people better than a 2K crop. But being the self-serving guy that I am, I'm not willing to give up the possibility of 2K scaled RAW if I have to sacrifice it in favor of 1080p. :biggrin:

Didn't we just have this discussion last friday?

[Poi Boy]

100% agree with you Jeff.
Aloha
-A

[Häakon]

I will definitely finish 2K and only do a scale to 1080p when mastering for BluRay, HD-DVD or broadcast.

I do agree with the need for a 1080p direct scaled option though. It would probably serve many people better than a 2K crop.

Didn't we just have this discussion last friday?

Brook closed it because some people derailed the topic onto something else. :bleh:

I still maintain that WAY more people will be finishing to 1080 for HD release of their material (whether it's EFP OR DCP stuff) than 2K. Please understand that I shoot features! I am not on a crusade for the corporate industrial shooter! I think people have it in their heads that "1080 equals HD, which equals video" and "2K equals cinema, which is clearly superior!" This is rubbish. Look at where your material is going.

Even if you are fortunate to have your project projected and screened, there's only one way to distribute it - and it 'aint 2K! DI becomes irrelevant with RED because you're already shooting digitally. And cinema projects will (should) be shot with the superior 4K anyway. The point is, even if you project your work at 2K, the only way to get it into the hands of the people is with 1080. That's where your material is going to live and breathe. And I would think that most people would want the highest quality material to get out there.

Understand that if 2K was a delivery format, this wouldn't be an issue. But scaling to 1080 in camera is going to produce a better picture than a double-scaled one, and what I'm not wanting to do is trade in a superior 1080 picture for a slightly larger 2K frame that very few people will ever see.

I guess I'm just surprised at people's reluctance to shoot in 1080 when quality is allegedly the motivation for keeping 2K around in the first place. Either I'm just not conveying the message properly (highly possible), or people are just blinded by the "2K" number and are willing to sacrifice their picture to get there. It's a little baffling to me, but I understand that I am in the minority on this one and completely respect your opinions and wishes for RED ONE.

[Alexander Nikishin]

I'll be finishing in 2K as well, I love me some Christie 2K!

[Häakon]

I'll be finishing in 2K as well, I love me some Christie 2K!
Right. Well I'll keep a lookout for all of your Christie screenings.

In the meantime, when a producer in Chicago wants to see your work, you'll be sending it to him on Blu-ray - and it's going to be a compromised picture because you've been championing for ditching 1080 this whole time. Did you even look at the candybar test?

[Poi Boy]

I"m pretty sure we are all going to get our way. Having said that Haakon, I think you overestimate the res hit from cropping the 2k scale and underestimate the huge advantage of the raw file that you will get regardless of how well lit or balanced you are at origin.
Aloha
-A

[Alexander Nikishin]

Would it even be possible to have 4K / 2K @100fps / and 1080P RGB firmware?

[Alexander Nikishin]

Right. Well I'll keep a lookout for all of your Christie screenings.

Well with your attitude towards me, you won't be finding an invite in your mail anytime soon. :bleh:

[Häakon]

I"m pretty sure we are all going to get our way. Having said that Haakon, I think you overestimate the res hit from cropping the 2k scale and underestimate the huge advantage of the raw file that you will get regardless of how well lit or balanced you are at origin.
Aloha
-A
Hi,

I certainly hope there is a solution for all. :-)

I wasn't ever thinking of cropping my image - if you're willing to lose those extra pixels, you should be shooting in 1080 anyway. The quality hit goes from scaling a 2K picture to a 1080 one, which is only 6% different in size. At that subtle of a resize, the picture is softened and resolution is lost.

Jim said that the scaled formats are RAW now, so it's not a RGB vs. windowed scenario. I am talking about the difference between scaled 2K and scaled 1080, not the 16mm options (which deliver more framerates). All of this is RAW, so there's no sacrificing of that workflow taking place.

[Häakon]

Well with your attitude towards me, you won't be finding an invite in your mail anytime soon. :bleh:

I'm just saying, what's the realistic outlook for the material that you shoot - that everything you do is going to be viewed on a Christie 2K projector? And by how many people? Way more people will be viewing/appreciating/critiquing your work on a Hi-Def disc format, so it makes sense that you ensure that picture looks as best as possible. I assume that you'll be shooting mostly 4K anyway if you're doing feature work - and in a 4K timeline, 1080 and 2K are going to look nearly identical.

[Poi Boy]

Haakon I think you may have misunderstood; You want the 4k down sample to 1080 and I'm happy with down sample from 4k to 2K. You believe there is a big resolution hit in 2k to 1080, I don't think that will be the case.
As long as they both stay raw I'll be happy with either.
Aloha
-A

[Alexander Nikishin]

I'm just saying, what's the realistic outlook for the material that you shoot - that everything you do is going to be viewed on a Christie 2K projector? And by how many people? Way more people will be viewing/appreciating/critiquing your work on a Hi-Def disc format, so it makes sense that you ensure that picture looks as best as possible. I assume that you'll be shooting mostly 4K anyway if you're doing feature work - and in a 4K timeline, 1080 and 2K are going to look nearly identical.

The realistic outlook is that I plan to capture in the highest quality possible with the mindset that ALL of my feature work will be shown in a 2K theater.

Whether that will come to fruition or not is yet to be seen, but I do not have the mindset that my films will be straight to Blu Ray titles.

If that were my goal, I'd be all for 1080P and I'd also be doubting my talent, which I refuse to do.

My goal is to shoot 4K for anything from 18-26fps, 2K for any slow mo up to 60fps and 2K windowed for anything 60-100fps with the end goal of the highest quality projection the budget for the shoot can afford.

[Michael Brennan]

I'm with Haakon on this one, not necessarily based on my personal needs but guessing at Red user profiles and the worldwide market for High Fidelity moving images.

There are tens of thousands of HD edit suites and decks out there that are not going to suddenly stop being used because RED exists.
Folks have to get a return on their HD kit.

If RED can also slip into existing HD workflows RED users will win work without threatening folks who have an investment in HD kit.
The more evangelical RED users are the bigger the barricade they will face.

The transition of RED being accepted universally with open arms isn't going to happen overnight.

Mike Brennan

[Jeff Kilgroe]

I guess the way I'm looking at it is many, if not most, large budget and Hollywood features are finished at 2K these days. They later make the transition to HD 1080p disc formats just fine and I haven't noticed softening being a huge problem. I will reserve judgement on this until it can be properly tested with data from a RED camera.

But I still want to say what I said in one of those other threads. Yes, I do agree with you Haakon, having the camera scale directly to 1080p and not 2K makes more sense for most people in most situations. If the scaled modes in fact will be scaled RAW data, then I can see a strong possibility that both 1080p and 2K scaled modes can be provided and that would be the best of both worlds. Like I said, I'm just saying what I intend to use most of the time.

If Jim and the RED team decide to go with 1080p scaled over 2K because it will ultimately serve more users, then I'm fine with that.

While we're on the subject and talking about a softening and potential resolution loss from scaling the size down 12.1% (2048x1152 is 12.1% larger than 1920x1080), I've got a few other things to think about. There's the same potential for softening in an up-rez of 1080p to 2K. It's more about how the scale up or down is done and not the scaling itself or if the image is scaled multiple times. And due to the way bayer data is arranged and constructed, it may be far more efficient to do a 1/4 scale to 2K than to something that's more arbitrary (1080p). We may get a superior image and more accurate color data out of a scale to 2K, thus giving superior results if the scale to 1080p is handled in post.

Having 2K as the recording mode and then doing a crop to 1080p may work fine in a lot of situations, provided proper framing guides are in place in the camera. And I see no reason they shouldn't be. Even if a 1080p scaled mode is provided, I could see some potential for shooting 2K and cropping -- there may be some unforeseen advantages that offset the loss of FOV. Like getting a 3% overscan on all sides of the image. This can help in smoothing out minor camera bumps, possibly some other things.

Anyway, that's all I have on this and don't want to argue it any further. I seriously hope that scaled RAW can be provided by the camera at BOTH 2K AND 1080P.

[Alexander Nikishin]

I seriously hope that scaled RAW can be provided by the camera at BOTH 2K AND 1080P.

Same here, I wish we never had to duke it out over 2K and 1080P and that they could just both be in there, would've saved me alot of friendships. :cold:

[Vincent Rice]

If you crop your image, you are losing FOV and wasting sensor resolution. Why not just downscale to 1080 and skip a step?

Because I like having the extra pixels available in post for the same reasons as Graeme suggested.

Like everyone else here I find your fixation with going straight to 1080 baffling.

[Häakon]

I think we've beaten this one to a pulp. It's clear we need to hear from RED before proceeding.