Why not 2k?

[Zach Tucker]

I have a question from a practical visual effects/post production/delivery perspective. For over a decade I have worked on many obnoxiously big budget hollywood popcorn movies. About 98% of the footage I've put my digital hands on has originated from 35mm film. True film res is clearly higher than 2k but for 100% of those projects we scanned the film at 2k, did all our CG and compositing at 2k and delivered at 2k. I know there is justification for 3k, 4k, even 6k in a true "DSMC" rig, but how many companies have the infrastructure to work with 4k footage at any reasonable scale? The biggest VFX/post facilities I've been at have over 3,000 processors to render on, but jumping from 2k to 4k would QUADRUPLE render time. It would quadruple disk space, RAM and network traffic. And due to the need to have 12,000 processors in this case (to render the same number of shots at twice the resolution in the same amount of time) it would probably require 4x the AC to keep those machines cool. So 8x the electricity to cover the additional machines and AC? LA has already turned down vfx facilities requests for more electricity.

So who is actually shooting 4k, working at 4k and delivering at 4k? (vote in my poll if you'd like) Throw 3d/stereo into the mix (something that is clearly happening more in the real world than increased delivery resolutions) and you can double the above numbers. Until someone makes a 64-core machine with 128gb ram and uses 500w for $3k, there isn't a facility in the known universe that could handle such a 3d project.

I've also worked on some high-budget commercials shot on the Red One. Again, the Red footage is only ever used at 2k and delivered at HD. On set when a higher frame rate was needed the image was cropped, effecting lens choices. I understand the limitations of throughput and heat inside the units. I also understand scarlet/epic might shoot higher frame rates without a need for cropping. It's all a long-winded lead up to...

How many film/commercial productions need the "S" in "DSMC" (ie, 3k-6k)?

Is the proud new owner of a scarlet going to deliver his independent film to the festivals at 3k?

What if there were a brain/sensor option in the upcoming scarlet/epic to take advantage of the massive global industry who is working at (and limited to) 2k?

If that sensor were the same size (s35/red one) but was only a 2k chip (compared to 4k) wouldn't each pixel be 4x the size and thus gobble 4x more light?

Would your storage go 4x as far?

Would you get more latitude?

Would rolling shutter issues (which are a NIGHTMARE for vfx) be easier to eliminate due to fewer pixels on the chip?

Would your encoding go 4x faster?

Would you be able to shoot 4x the frame rate with the same horsepower and no need for a sensor crop?

It's kind of like having a Ferrari in LA. Sure, it *can* go 200 mph, but even when you're not crawling in traffic the cops are gunna nab you at 80 mph.

.z

 

[Milan Nikolic]

All good questions!

 

[Sven Seynaeve]

That's what keeps us up all night.

Wish I had the answers instead of the knowledge and awareness..

 

[Stefan Christou]

The price of computing power halves every two years. 4K is the future, or perhaps just recent history.

 

[Dan Hudgins]

Bayer vs. RGB

Bayer vs. RGB

A 2K Bayer camera with OLPF does not give the same 2K resolution as a film scanner with a filter wheel shooting RGB into a monochrome 2K digital camera.

You need to shoot 4K in a Bayer camera to get clean 2K images RGB for post because of the losses in the sensor, de-mosaic (de-Bayer), low pass to remove chroma moire, OLPF, high speed camera lens, etc.

Also you need a good image down sizer to reduce the 4K camera images to 2K to avoid aliasing.

4k helps hide matte lines better even if you downsize for filmout.

My system can do a 4K post for low budget films, how many PCs you need depends on how many films you need to make per year, but it can be done at home with several PCs in about the same time it would take you to do a feature with 35mm mag-film and to cut the negative yourself. A good feature post takes a year or two to compose the music, do the ADR and foley, grade 800+ shots, edit, and the many other details anyway, so its not such a big deal to have 10 basic PC running FreeDOS in another room, the render computers do not need a monitor or keyboard or mouse, you can use a switch box to setup each batch process, and use removable harddrives to work on each reel.

 

[Zach Tucker]

A 2K Bayer camera with OLPF does not give the same 2K resolution as a film scanner with a filter wheel shooting RGB into a monochrome 2K digital camera.

You need to shoot 4K in a Bayer camera to get clean 2K images RGB for post because of the losses in the sensor, de-mosaic (de-Bayer), low pass to remove chroma moire, OLPF, high speed camera lens, etc.

Also you need a good image down sizer to reduce the 4K camera images to 2K to avoid aliasing.

So would this suggest that the 4k footage (off of a 4k Bayer camera) would exhibit these same mosaic/moire/aliased properties?

If the answer is no, what is being done to the 1:1 4k footage to prevent it? And couldn't the same thing be done to a 2k sensor with 1:1 2k footage?

.z

 

[Evin Grant]

Oversampling is your friend...
For many reasons already stated but primarily because even if your final product was 720P or SD you will realize more precision and thus a superior image by shooting 4K. I doubt I will finish a project in 4K in the next year or two but acquiring in 4K (or more) has many advantages when it comes to the final 2K image. First off the measurable image detail of a 4K Red file is 3.2K. This is a factor of the bayer pattern and the anti-alias filter. This is still quite a lot of resolution and produces a stunning image projected at 4K. For 2K projection it offers the ability to stabilize or push into a frame without sacrificing quality as well as the overall better tonal range and precision that a down-sampled raster affords. If you think oversampling has no effect on final image quality watch the newly re-mastered Baraka Blu-ray on a really good 1080P projector. That film was scanned at 8K from the 65mm negative as well as mastered at 8K and then down-sampled and encoded to Blu-ray. It's stunning and is clearly superior to normal 35mm/2K scanned films. Of course not all projects deserve this treatment but it makes a very compelling argument to shoot at least 1-2K more than your finishing format.

 

[Charles Angus]

One word:

Flexibility

 

[Peter Majtan]

Bayer and oversampling are all fine, but I believe there is a limit. Given the 2K deliverable You only really need 3K acquisition. Now Jim can attest to how long I have been advocating (and pulling his strings... :sifone: ) a S35/FF35 sensor with 3K resolution for all the reasons stated above. For a while he thought that I meant 3K in USD and he was quite upset about it, but once he realized I was asking about 3K resolution he asked something like "But why would You want less resolution?" - and I did present pretty much the same arguments as above. I am glad I am not the only one thinking along those lines to be honest. Especially with the current state of economy - the resources required for true 4K DI workflows are out of reach for everyone, but large facilities with reasonable budgets. I am trying to do my bit by helping storage manufacturers to design and develop high-end storage solutions that won't break the bank (I brought one of them here recently as sponsor...). And the new Mac Pro's and XSERVE's are certainly doing their part along with the soon to be released Snow Leopard and Final Cut Studio 3...

But in the end I would love to see S35/FF35 Scarlet with 3K (resolution) sensor. And I would pay for it the same money as for the 5K model without hesitation...

Just my two cents...

Peter

 

[Alex Carr]

I am compositing in After Effects in 4k, but at a 2k sequence for delivery. As I cache the preview to my Hard Drive it becomes much easier to work with... But the rendering is astronomical... 1-1/2 hours for a 2-1/2 minute scene without major compositing...

 

[Chris Swinbanks]

Regardless of the capture medium (which has been mostly 35mm film for us up until Knowing) pretty much everything that I've worked on in digital for a film output in the last ~ 14 years has been 2k output resolution or less, the same as Zach states (eg, Matrix, Moulin Rouge, Babe, etc etc).
Like other vfx facilities, having film scanned at 4k or more res (where warranted) is done for the sake of the fx or to maximise the clarity of the down-sample, but the bulk of the work after the capture/transform stage is 2k or some subset of that.
An effects heavy shot can have an enormous number of layers going together to create the composite.... vastly increasing the storage/bandwith/ backups etc required to ensure the useability and safety of the media.
Knowing, while "captured" at 4k (I'm ignoring the current debate in CML :Yawn was immediately resized to 2k for the vast majority of fx, and the DI.

In the past, some facilities have "sold" 4k as the only way to go, which I believe has been a bit of a scare tactic.
Does 4k on film look nicer on screen... yes, a smidgeon, by the time its gone through I/P - Dupe Neg - Print.... but for the majority of product out there, not worth the overhead in post, until computing power and composite/grade/playback systems can reduce these overheads to equate to how we deal with 2k now.

just another 2c......

and happy easter all!
Chris

 

[Deanan]

Like other vfx facilities, having film scanned at 4k or more res (where warranted) is done for the sake of the fx or to maximise the clarity of the down-sample, but the bulk of the work after the capture/transform stage is 2k or some subset of that.
An effects heavy shot can have an enormous number of layers going together to create the composite.... vastly increasing the storage/bandwith/ backups etc required to ensure the useability and safety of the media.
Knowing, while "captured" at 4k (I'm ignoring the current debate in CML :Yawn was immediately resized to 2k for the vast majority of fx, and the DI.

I'm looking forward to direct R3D support is in vfx compositing sw as there will be considerable savings for storage/bandwidth for source plates. Rather than pushing 9MB or 35MB/fr around for 2k/4k you're pushing only ~1.5 MB/fr. There's also the nice advantage of being able to decode at various resolutions as you need in the comp tree.

For certain things like roto, keying, some types of filtering, etc, you do gain by taking advantage of the extra pixels before down sampling while other steps like color processing and layering are better off after the downsample.

Theres now downside to starting off at a higher resolution especially if you have the choice to down sample where you need it.

d

 

[Chris Swinbanks]

storage of all the other elements and finals in redcode could be very nice too.... nice of you to offer it up Deanan!!! :rofl:

 

[Bob Torrance]

4K

4K

If you want the depth of field of 35mm on Red you need to capture at 4K. More lens flexibility. Of course uncompressed is a different matter. To me things seems to be moving to 4K as a new batch of Digital Cinema Theaters are starting to use the Sony 4K projector. Maybe the new redcodes (250, etc) will be helpful for a 4K pipeline.

Bob Torrance

 

[Dan Hudgins]

2K vs. 4K camera, is 3K post optimum

2K vs. 4K camera, is 3K post optimum

So would this suggest that the 4k footage (off of a 4k Bayer camera) would exhibit these same mosaic/moire/aliased properties?

The 4K camera and its de-moasic use low pass filters to reduce the chroma moire, alasing and in the case of REDONE the wavelet related artifacts.

If you use a 2K Bayer camera those artifacts cover four times the image area and begin to be large enough to see in the result, so you are doing better to use a 4K camera Bayer camera than a 2K Bayer camera.

A monochrome camera in a filter wheel film scanner does not require interpolation to fill in the missing pixels in each color, and does not use REDCODE so does not require low pass filtering of the chroma data. The scanner and camera lens and film limit the high frequencies so the OLPF can be ommited or at least have a smaller radius of blur because the monochrome pixels are closer together in a sence since they but side by side, whereas in the Bayer camera the red and blue pixels do not but side by side, nor do the green.

RED Co. decided not to make a three chip camera using relay optics and three monochrome sensors because it would be very large and heavy, and cost 20x as much to market, despite what they have said, maybe because they do not understand how to do the optics right, the results could be better both in the quality of the color and the resolution if recorded RAW, if done right with the light through the prisms being only parallel rays.

REDONE does not record raw, and the Bayer filter requires low pass filtering, so they were right to use a 4K sensor since having the higher resolution reduces the size of the artifacts and gets an end result at 2K better than you would get with all things equal in a 2K camera.

Analog video can resolve fractional pixels, like 1, 1.25, 1.33, 1.5, 1.75, but in digital images you just get 1 or 2 so much of the high frequency detail that is smaller than 2 pixels is lost or distorted.

Using 4K gets you finer high frequency detail above the 2K limit, closer to analog video and film rather than 2K digital images.

Would 3K be enough to get all you need above 2K, maybe if the camera recorded RAW it would not need as many low pass filters as one using REDCODE. I would say that a 3K S35 sensor camera recording full RAW data using true lossless compression (Delta Modulation) would give maybe better results than the REDONE. It would be easer today to build an uncompressed 3K camera than it was when REDONE was desigined years ago.

3K is managable from a post standpoint, if you are doing 3K 16:9 you can use files compressed with lossless compression to save disk space.

My system can post 3K at any aspect ratio, or 4K or larger. Saying that doing a 4K DI is not something that is an option is not true, my system is desigined to allow filmmakers to do their DI at 4K if they want, thats why I have written it, mostly since we were going to work with high resolution film scans, and all un-compressed since we remember what 35mm prints looked like before there were filmmouts and no projectionists. Degrading the theatre experence by going digital over film-to-film was NOT OUR goal, better was.

Doing Post at 3K, then going to a 2K at final filmout or projection should result in better quality if done right. You should avoid programs that do crude sub-sample to resize since that does not give the best results, I am writing code into my system now to do a better job on the down size and the results look quite good so far, it should be out soon.

2K filmout does not equal what 35mm film to film printing can produce under the best circumstances, but without a projectionist using a spoting scope to keep the images in focus you can never see what is on the film up to its limit since the projector is out of focus. Rather and argue about if you can see any thing better in a 4K filmout over a 2K filmout, we should be developing some way to keep the projector in focus. I think a remote control box could be given to someone in the audience that runs a servo motor on the projector lens, and they can turn the knob on the remote control box while they watch the movie to keep it in focus, that would not cost too much to install and would do more to improve the resolution in the Cinema than any 28K compressed digital cinema camera.

I was watching CHE #1 and #2 to see the 4K quality of the REDONE, the projector drifted in and out of focus, mostly out of focus. Later to find out that it was it seems a 2K print made on a Cinevator. Since the projector could not even hold focus at 2K I can see why people question the need for 4K filmout. Well bring back 70mm prints and a projectionist and then you can see maybe what was shot at 4K.

 

[Hrvoje Simic]

Until someone makes a 64-core machine with 128gb ram and uses 500w for $3k, there isn't a facility in the known universe that could handle such a 3d project.

One more reason to start harvesting this.

http://www.nvidia.com/object/GPU_Computing.html

 

[Jeff Kilgroe]

The biggest VFX/post facilities I've been at have over 3,000 processors to render on, but jumping from 2k to 4k would QUADRUPLE render time. It would quadruple disk space, RAM and network traffic. And due to the need to have 12,000 processors in this case (to render the same number of shots at twice the resolution in the same amount of time) it would probably require 4x the AC to keep those machines cool.

I know where you're coming from with this, but we also have to consider the evolution of those processors, storage systems, etc.. Processors today are 10% cheaper, consume 20% less power and are 4X as powerful compared to 4 years ago. Look at the farm of Pentium and Pentium-II CPUs that WETA used for LOTR FX work about 10 years ago vs. what is available today. Hard drives have increased in capacity and decreased in price per GB by even larger margins. Storage capacity these days is mostly trivial.. The real puzzle within the storage equation is proper back-up, archival and maintenance of the data.

In a couple more years, processing, posting and delivering 4K will be no big deal. It will be easier than doing the same with 2K only a few years ago. Combine that with native support for R3D as Deanan and others have mentioned. That right there already demands a fraction of the storage requirements and bandwidth vs. 2K DPX. If post systems can work with native R3D wavelet data and decompress required frames at local workstations and render nodes on the fly, that is one more step toward making full use of 4K acquisitions and beyond.

IMO, I think the biggest hurdle left for 4K post is not storage or processing power. It's economical display solutions. Give me an affordable way to display a 4K image on a 50" or smaller display with proper color space and I'll build an edit / color / compositing suite to use it, even if I can't quite run 4K debayer in real-time for another iteration or two of CPUs.

 

[Gavin Greenwalt]

I know the owner of Splutterfish (Now owned by Caustic) and they're releasing their CausticOne card this year which will deliver 20x raytracing accellerating followed by 200x raytracing next year. How? Hardware specific solutions.

If the market dictates that there needs to be 4k debayering in real-time you can bet your bottom dollar that someone will release a small relatively inexpensive chip which does it for less than $2k.

We're already seeing GPGPUs which unlike raytracing are actually an excellent chip architecture for debayering type algorithsm. And modern GPUs have hundreds of cores. I just saw a H264 encoder today on Gizmodo that's just a USB dongle and costs <$100.

If people want 4k... 4k won't be difficult to deliver. The problem is right now the delivery system I think more than the hardware to run it.

If I were to pick a technology I want to depend on advancing in performance it's the processing front. The storage front on the other hand pokes along at a snails pace. That might be changing with SSDs... but historically MB/s has lagged behind processing power.

VFX people are used to working offline/proxy. We're used to dealing with nowhere near real-time and only just now starting to get a taste of it with a swath of new RT/Near RT 3D Solutions and hardware accellerated compositing. I wouldn't worry too much about 4k on that front. Work with the proxy. Work with Domain of Interest. Then go to full rez and render overnight on the farm.

Even if 4k is 1/20thRT it's easily within reach for a small studio to own a 20 system renderfarm for rendering out the Online Edit.

Desktop 3D in near real-time is just around the corner.

I don't think performance is something we need to worry about too much.

 

[Dan Hudgins]

Exactly

Exactly

Even if 4k is 1/20thRT it's easily within reach for a small studio to own a 20 system renderfarm for rendering out the Online Edit.

Exactly. 20 PC running FreeDOS without monitors, keyboards, or mice do not cost that much, just a motherboard and removable harddrive, and you get near 100% CPU time (?).

 

[Häakon]

There are two factors to consider when shooting in high resolution: acquisition and delivery. The original poster is probably right that the number of projects actually delivering and showcasing in resolutions higher than 2K are in the vast minority. But the reasons for having a high-res acquisition system are well documented (and echoed by many posts in this thread).

When it comes to a 1080/2K delivery, 3K RED is as low as I ever want to be acquiring, and in my completely subjective opinion, even 4K isn't enough. If you've ever intercut 2K RED footage with 4K shots - even if you're aiming for a 720p web release - you can very easily tell the difference. I can't wait to get my hands on a 5K Epic/Scarlet for this very reason. Of course, there's a limit to how many pixels you can comfortably cram onto a sensor before the disadvantages start to outweigh the benefits, which is why sensor size increases are going to start playing as important a role as megapixels down the line. We've already seen this time and time again in the digital still world. Fortunately for us, RED already has large-sensor cameras on their roadmap.

Anyway, the point is that acquiring images in as high resolution as possible is almost always preferable if the option exists regardless of whether or not you're aiming to deliver at a cinema, on Blu-ray or plain vanilla DVD. Oversampling is king.