In the edit bay, stereoscopic content can be output from a workstation in a number of forms.  AJA and NVidia can output separate SDI streams for each eye, using the same display interface that would be used onset.  Combining the left and right view into a single stream allows it to be played out a single SDI port or even standard DVI on a GPU to a passive 3D display.  Those specific options will depend on what software you are using, and need to match your display requirements.  Outputting 120hz from a dedicated HDMI 1.4 port requires a newer GPU and specific application support, but greatly simplifies the hardware setup.  I am sure there will be a whole variety of new options announced at NAB next month, and as new hardware and software tools are developed, the preview options become cheaper to setup and easier to use.

Shutter glasses can be used with displays that have high refresh rates, to filter alternating frames from one eye or the other.  A 120hz display can provide 60 frames to each eye, which is enough to create smooth motion and make the blinking shutter imperceivable.  The disadvantage is that each viewer requires a pair of expensive active shutter glasses, which all require charged batteries or some other source of power.

Polarization has been used in many different variations of 3D display solution.  The most straightforward is to use dual projectors, each projecting onto the same silver matte screen, with opposite polarizing filters in front of the lenses.  Complementary filters are used in cheap passive 3D glasses that each viewer wears, filtering out light from one projector or the other, from each eye.  One issue with this method is that perfectly aligning two projectors on a large screen can be very challenging.  To solve this, RealD took this process a step further, and created a polarizer for a single projector that alternates the polarization angle at the same rate that shutter glasses usually blink.  The result is that the “active” part of the filtering for a 120hz projection is done once, and each viewer can use cheap passive 3D glasses to separate the views for each eye, while experiencing the full resolution results offered by shutter glasses.

Polarization can also be used within passive 3D LCD displays, but currently with these solutions, any pixel is permanently polarized either right or left.  As long as the image sent to the screen conforms to the pattern, usually horizontal interlaced, it can send imagery from the correct angle to each individual pixel.  This requires a 1:1 image to pixel ratio, in order to ensure that the display places pixels from the correct angle in the correct locations.  The disadvantage is that usually half of the image resolution to each eye is lost to the filtering process.  Displays that have this type of built in polarization can be viewed with cheaper passive 3D glasses.  In my opinion, any passive 3D is also easier on the viewer’s eyes, especially over periods of extended use, since there is no alternating flicker.

The cheapest, but lowest quality option for viewing stereoscopic depth information on regular displays, involves using color information to separate the images for each eye.  This is called anaglyphic processing, and the benefit is that nearly any standard full color display device can be used, but it comes at the cost of most or all of the original color information being discarded.  Anaglyphic video can either be generated on the fly during playback from two separate streams, or rendered out into a single stream to be edited, usually as an offline version.  Viewing anaglyphic imagery, or even just wearing anaglyphic glasses for any significant period of time is usually not comfortable.

On the other end of the spectrum, there are certain displays available that use lenticular filters to isolate the separate views, and focus them on each eye.  This is similar to the process used to create depth and motion effects on static 2D surfaces, called lenticular printing, but using this method for video requires the underlying display to be extremely high resolution.  With stereoscopic source, this results in certain areas where the effect is null or reversed, and requires the viewer to be in just the right spot for maximum effectiveness.

The next step to solve that issue is to replace the alternating left and right views projected from the lenticular screen, with a series of progressively different views.  That way any viewing angle within reason can be accommodated for, without gaps where the illusion breaks down.  Instead of recording every variation of viewing angle, these perspectives are usually generated on the fly from a single recorded perspective, usually with the help of a Z-depth map.  The Z-depth channel is like an alpha channel, but instead of storing values of transparency info, it records how far an object on screen is from the camera.  This information can either be captured at record time with specialized hardware, or be generated from an image analysis process, with a bit of interpolation.

At the end of the day, each option has certain pros and cons, and multiple display types could be used at different stages of a single project.  As long as you have both full streams of data available, you can adapt your content for any display.  Once you render it into a display specific format, like interlaced or anaglyphic, certain image content may be lost that cannot be recovered for use on other  display types, without going back to your original source.  Because of this, it is highly preferable to have a system that allows you to display stereoscopic content in 3D without having to pre-render into a dedicated display format.  As computers become more powerful, and better software is developed, that capability is becoming much more common, greatly aiding the stereoscopic post process.

The first step in the process is to generate and record two separate images of similar but different perspectives.  This is usually solved with specialized hardware, by using rigs to mount two narrow cameras in parallel, or using a beam splitter to record two similar perspectives with larger cameras.  Recently there have also been cameras released that have multiple lenses as an all-in-one option.  Ideally both the separation between the lenses and the angle at which they converge should be easily controllable, especially for live recording, since those two values have a significant impact on the final stereoscopic effect.  The other acquisition option for content generation, is to render two separate perspectives of a computer generated 3D animation.  This requires no special hardware, just a few changes to the software and settings, plus a doubling of render time and storage  space.

Depending on the acquisition approach, there are a number of extra steps in the post process.  Both angles must be matched together, synced down to the frame, and usually muxed in some fashion.  Eventually they need to be aligned, the convergence needs to be set, and there may be a color shift from the beam splitter that needs to be compensated for.  Any visual effects will be much more challenging as well.

Finally there is the issue of displaying separate images to each eye from the viewers perspective.  This can be achieved with separate miniature screens for each eye, or more frequently, by finding some way to filter out the opposing view from each eye, from the combined image.  This can be done with color filters, polarization, alternating shutters, or lenticular screens, to control the image that each eye sees.  Ideally there need to be ways for the stereoscopic effect to be previewed immediately on set, examined during the post process in the edit rooms, and delivered to the final target audience.  This makes display solutions the first logical aspect to be examined, even before camera options.  The basic ideas behind many of these viewing options will be discussed in the next post.  Following that, we will examine some of the other practical ramifications imposed on the post-production process by stereoscopic projects.

Shared SANs takes the benefits of having all of your storage interconnected with high bandwidth links, and extends it one step further.  By running special software to synchronize the connected systems, it allows each of the connected systems to access the same the data on the same volume on the SAN, without overwriting each others files or corrupting the data.  Most SAN software is designed to function as a peer to peer solution for smaller installations, (5-10 systems) or with dedicated servers for larger SANs.

As is probably obvious, there are many benefits to having multiple systems sharing the same set of files on a central high performance disk array.  First off, you don’t have to buy individual arrays for each system, making individual systems cheaper and quieter.  All the actual data is stored in single physical location, making it easier to protect and secure it.  With all the data stored on centralized volumes, file management is easier, with a single unified file structure, and you lose the need to duplicate source files across every system that needs local access to them.  This saves time and storage space.  It also makes it easier to make thorough backups, especially in automated form, which makes your data more secure.  On the flipside, the initial investment is usually rather high, and all of your eggs are in one basket.  If the SAN has an issue or problem, your entire production may grind to a halt until the issue is resolved.

Most all SANs use Fiber Channel as their primary physical interface.  Although this in not inherently required, until recently there was no other standard technology that offered that capability.  CalDigit recently launched a PCIe switch product that they claim offers shared SAN capabilities for their PCIe attached arrays.  While the idea is great, currently the hardware is still in a similar price range to entry level fiber solutions, and you still need expensive software to keep the connected systems in sync and prevent your SAN data  from getting corrupted.

iSCSI also offers some of the same capabilities, with block level drive access, but is only a viable competitor in the high end production world when running on 10Gb ethernet interfaces, which are still usually prohibitively expensive at this point.  Running iSCSI over Gigabit ethernet may be a viable solution for certain compressed workflows, but offers few advantages over regular network storage, at the expense of needing separate SAN software to share properly.

There are a number of different software options when creating a Shared SAN.  I am not familiar with every one of them, but the five I describe here should give you a place to start.  They all serve the same purpose of preventing multiple systems from trying to write data in the same spot at the same time, but they use a variety of different methods to accomplish that objective.

FiberJet is the cheapest option, but does not allow true file level sharing.  It prevents overwriting and data corruption by only giving one system at a time write access to any given volume.  On the otherhand, all systems can be given full read access any volume all the time.  This allows you to share source footage and other media with multiple workstations without the waste of having to duplicate the files.  It doesn’t allow you to easily share actually project files, since most apps will require write access, and will usually force you to share your files across a number of separate volumes, making it harder to find or backup your data efficiently.  So FibreJet gives you about half of the benefits of a Shared SAN, as a low cost starting point.

MetaSAN has been available for quite a while now, and is fairly common in PC based post-production environments.  It supports true file level sharing, allowing all of your systems to read and write files on the same volume simultaneously.  It supports standard file systems, and operates as a separate process over IP to keep machines in sync.  It also allows PCs to access files on Mac formatted drives and vice versa.  It requires one of the connected systems to host the server process, to manage the distribution of metadata and synchronization information.  That system does not have to be dedicated to that task, but it can be for maximum performance and stability.  If you use a user workstation, rebooting that system could cause other users to lose disk access.  I have used MetaSAN for many years, and it is an amazing tool, but it has its quirks that you have to get used to.  It has a tendency to freeze up workstations if something goes wrong, as it waits for certain requests to timeout, which can make it difficult to troubleshoot when you are in a hurry. (And when the SAN is down, you are always in a hurry)  On the otherhand, with all of its instability and frusteration, it has never allowed one of my arrays to become corrupted, or for me to lose data, so it clearly performs its function.

HyperFS is a recently released option, primarily offered by Rorke Data in the US.  It has its own proprietary file system, which can be directly accessed from Windows, OSX and Linux based systems.  The base software is priced similar to MetaSAN, and functions in a peer to peer fashion in smaller installations.  But if you have more than 8 systems to connect, you will be required to invest in a full dedicated metadata server and license, which significantly increases the deployment cost.

XSAN is Apple’s shared SAN software offering, currently on version 2.2, and it is limited to OSX and requires Xserve systems as metadata controllers.  As a PC guy, I have no experience with XSan, but it is used by many Mac based post-production facilities.  The underlying technology is based on the last option we will examine, StorNext.

StorNext is by far the most expensive option, but it offers higher performance, specifically for frame based media, than any of the other choices.  Frame sequence based media bog down other SAN software due to the high number of individual files that are being opened, accessed, and closed, in rapid sequence.  Each individual frame requires the same amount of metadata and synchronization data as an entire video file, overloading lower end software options.  StorNext is an enterprise level product with a variety of options and tiers, with versions that support every different OS, and even ones that interoperate with Apple’s XSan.  It is clearly an expensive option, but you are paying for stability and performance, putting it at the core of many DI facilities that have a DPX based workflow.

Shared SANs are one off the most complicated and expensive investments available in the post-production world.  Lower cost network based alternatives are a better place to start, for smaller oragnizations and compressed workflows, until you are sure you need the performance that SANs can offer.  Once you are working with uncompressed high definition video, or 2K frame sizes, especially with multiple users, a SAN will probably be worth the investment.  The effect that they can have on your workflow and level of collaboration is dramatic, making them worth the effort it takes to get them up and running.

External SAS connected arrays function in much the same way as SATA based ones, but with a few advantages, that usually come at a significantly higher cost.  SAS is a full duplex interface, and the command set is based on SCSI instead of IDE, allowing higher performance and throughput.  More expensive SAS arrays also support multipath signaling, for greater redundancy in the supporting electronics. (As opposed to the redundancy provided at the disk level by RAID configurations)  SAS also supports much longer cable lengths, up to 10 meters or 30 feet.  This can be advantagious for quiet video editing rooms, since the disk array, which is usually the loudest part of the system, can be located farther away from the users.

A number of vendors have now begun offering external arrays that interface with the host workstation via a direct extension of the PCIe bus.  This allows all of the RAID functionality to be contained within the array, and gives full speed access to the data as if it was contained within the machine.  Among the advantages of removing the RAID functionality from an internal add-on card, are that it can be attached to a laptop via an ExpressCard, which uses the same signaling protocol as PCIe, and that with addition of a few cheap pass-thru cards, an array can easily be moved between systems.  This is definitely not a hot swappable solution, since it accesses the PCIe bus directly, which is initialized at bootup on most systems.  But if your main edit system has a total OS meltdown at a critical point in your project, it should be much easier to access your data from a different system than if you needed to reinstall the PCI SATA RAID card somewhere else, and allow you use your laptop as a backup edit system in certain instances.

Fibre Channel is by far the most expensive option.  Every part of the system is more expensive, the PCIe HBA cards, the fiber cables, and the disk array controllers.  On the otherhand, Fibre Channel offers capabilities that none of the other storage options really do.  It is a hot swappable interface, running on fiber cables that can extend access thousands of feet if desired, and can easily be networked and shared.  Devices can be connected directly together, shared in an Arbitrated Loop, or all attached to a central fibre switch for simplified management.  It is an efficient and low latency interface, and is available in speeds of 1,2,4, or 8Gb per second, and multiple channels can be combined for higher performance.  Higher speed devices are usually backwards compatible with older hardware, similar to the way ethernet works, allowing you to upgrade your storage network one piece at a time.

Choosing the right storage solution depends on your immediate media needs, your available budget, and the direction you anticipate growing in the future. SATA based solutions offer all of the speed you could need if scaled large enough.  SAS can offer similar performance in a smaller package, but at a higher cost.  Sharing data beyond gigabit network speeds requires a storage system that can interface with multiple computers, but that comes at a significantly increased initial cost.  Investing in Fibre Channel storage is usually only worth the expense if you anticipate the need to share your data on a SAN, either immediately or at some point in the future.  I will examine a few popular shared SAN options in my next post.

AJA released a variety of new products, most of them adding 3D related features to previously existing offerings.  The Kona3G is a revision to the existing Kona3/Xena2Ke that adds stereoscopic support through HDMI 1.4 output and dual stream SDI-3G I/O, while also dropping in price about 30%.  Their Hi5-3D replaces the Hi5-3G and adds HDMI 1.4 output as well as a variety of options for processing dual stream and muxed stereo inputs.  The Ki Pro Mini is a smaller version of the Ki Pro that can now be mounted directly to camcorders, and record ProRes files directly to CompactFlash cards from HDMI or SDI inputs.

Blackmagic Design had their own selection of new products to announce.  Their line of SDI routers must have really taken off, because they are really scaling their offerings upwards, with new models that offer up to 288 Channels of input and output, and options for coaxial or fiber based connections.  I am just now upgrading to a 16×16 3G Micro VideoHub, but maybe someday.  Their other big news comes from their DaVinci line, with previously announced Resolve 7.0 for OSX finally being released.  It will be interesting to see what effect that new option has on the marketplace and the price of existing products.  My company will definitely be looking into setting up a Resolve system, and our primary colorist is very interested in the new capabilities it would give us.

NVidia released their new line of Fermi based Quadro graphics cards at Siggraph.  The Quadro 4000, 5000, and 6000 will be the successors to the current QuadroFX 3800, 4800, and 5800, with similar form factors and interfaces.  I have not had a chance to test them myself yet, but everything I hear has been positive.  Adobe has announced that the Quadro 4000 and 5000 will be officially supported for CUDA acceleration by the Mercury engine in Premiere Pro CS5.  This support came in the form of the Premiere Pro 5.0.2 update last week, which also adds the GeForce GTX 470 as a fully supported GPU option.  More significantly from a technology standpoint, they added 10bit Displayport support for Quadro cards, and also support for RedRocket acceleration, RMD files and newer Red camera updates, as well as better support for Broadcast Wave and certain XDCam-HD files.  There is also a 6 page list of smaller fixes in the new release, many of which fill significant holes in certain workflows.  I haven’t had much time recently to test out the new features, but getting 10bit color to my Dreamcolor is at the top of my list.

Avid Media Composer 5 was released three months ago, and they are now on the 5.0.3 revision.  I have had a chance to test it out at work, and while the ability to import any Quicktime via AMA can totally change the workflow for certain tasks, I wouldn’t use it as a primary way to edit large projects.  The performance and stability is not up to the same level as MXF based editing is.  You also need a very fast system for that to work.  Most of our Avids are HP XW8600s, while we have been dedicating our newer Z800 systems to CS5, but AMA playback requires more computing power than DNxHD editing, which is to be expected.  We also found Version 5 to be less stabile and less responsive on our large feature length DNxHD project, even without AMA based media.  The fact that it fully supports Windows 7 will be the factor that motivates our facility wide upgrade in the near future.

I was also able to test Media Composer 5 with my Matrox MXO2-Mini, for hardware HDMI out, and while it worked great at first, I once again saw a major hit in stability, with an escalating number of system crashes.  It is so close, but not quite there yet.  Hopefully we will see many of those issues worked out in intermediate dot releases, since most of these features are brand new.  Stay tuned, since I have a few other new toys that I will review in my next post.

FTC Disclosure: Many of the companies I refer to above have made their products available to me or my company in the past, but none of the new items discussed above were provided to me without independently purchasing them.

Adobe announced all of the features of the new highly anticipated Creative Suite CS5, which will include native 64bit, CUDA GPU acceleration, and better support for formats like DPX and DSLR clips.  I will have many more CS5 details in future posts, delving into how that will change the landscape of many workflows.  There are lots more new developments in the post world, but those are the primary things that stick out to me right now.  We still have two more days, and I will be spending a good bit of time demonstration Cineform’s Neo3D features at their booth at the back of the Lower South Hall.  So if anyone wants to catch up with me, feel free to stop by.

I know I haven’t posted anything here in a while, but there will be quite a bit coming in the near future.  I am writing a series of articles that will be posted between here and Shane Hurlbut’s blog at hurlbutvisuals.com.  I have worked with Shane on a number of projects, the largest one being the Navy Seal film over the last year.  I will be offering an overview of the post workflow options for video DSLRs on his site, with links back to this one to delve into the less glamorous, nitty-gritty details.  So stay tuned for some very focused articles detailing specific workflow obstacles in the next few weeks.  And then NAB is only a month away, which I am sure will bring a whole variety of new things to talk about.  I will be at Cineform’s booth as a demo artist/workflow consultant again this year, so feel free to stop by and check it out.