We've been doing nothing but live-action stereoscopic production since 1999. Our biggest challenge in the beginning was that we couldn't just go out and buy a 3D camera. We had to actually create the camera before we could do anything.
We started out trying to create our own camera from scratch. We literally bought photo sensors from Kodak, and lenses from Schneider, and hired an electronic engineer to design a camera board to control the whole thing.
At the time, MiniDV was a sort of industrial professional standard, so we wanted to make a camera that could record on to MiniDV at 24P, and with full color in every pixel. Our idea was to use a 1920x1080 Bayer-pattern sensor to create "super pixels:" combining individual red, blue and green pixels into a single pixel with full color information.
It quickly became apparent that it was a huge undertaking, beyond the scope of our budget and this one individual's capabilities. That's when we decided that it might be easier to take an existing camera and modify it.
While we were trying to work our way through all of these challenges, Panasonic came out with the AG-DVX100. I said, "Aha! Here's a camera with RGB data at every pixel, it records to MiniDV, it's small - this is going to be great!"
We got two of those, and said, "Now we want to do 3D - how do we get these cameras genlocked?" We went back to the electronic engineer we had worked with before. It took him a long time to figure it out, but he did. We now had genlocked DVX cameras, recording 3D to MiniDV.
That was fun for a while, but, obviously, MiniDV tape is very limited. The distance between the lenses -the interaxial spacing - was also pretty wide, which limited some of the things we could shoot.
As we set out to improve the camera, we ran into some guys at Purdue University who were working on an uncompressed interface for the DVX100. We helped with financial and logistics support, and we were the first to use the technology they created: a DVX100 with uncompressed 4:4:4, 10-bit RGB RAW output.
They also created a unique algorithm that allowed them to process that RAW sensor data into a 1280x720, 720P high-def image for each eye.
When we released it in 2004, we called it the 3DVX. At a hand-holdable 19 pounds, it's still the smallest and lightest self-contained, RAW-recording stereoscopic camera system in the world.
BUCKETS OF BITS
But with the improvements we added, the camera's data rate skyrocketed to 29 MB per second, per eye - not megaBITs, megaBYTES. That's almost 60 MB/second that needs to be stored in some way.
We couldn't record this copious data rate to MiniDV. We couldn't use P2. We wound up integrating a pair of stripped down Mac Minis - including CPUs, RAM, and the latest version of the Mac OS - to allow us to record straight to disk.
In the DVX100, the CCDs are caoturing 12 bits per pixel. We use user-defined lookup tables to record 10 of those 12 bits. There is a USB data interface inside our camera that connects directly to the camera's analog to digital converters. The 10 bits of data, just ones and zeroes, fly over this USB data interface and go directly to the Mac Minis, where it is then stored on removable hard drives.
(We record RAW data rather than a specific video format. The only application that recognizes the data is the application that was specifically written for the camera.)
Recording to hard drives also meant that we no longer need the VTR inside the cameras, so could get them much closer together. The interaxial spacing came down to 2.75 inches, a much more natural distance, more in keeping with the 2.5 inches between adult eyes.
Traditionally, we have been using high-speed 7200 RPM drives with a specially-selected case and interface that allow the drives to maintain high data rates. Hard drives have been problematic for a number of shoots, most particularly "Call of the Wild," which we shot on location in Montana in February 2008.
As you might imagine, it's pretty freaking cold in Montana in February. There were blizzards. Hard drives don't like that. They are rated to operate between 45 and 105 degrees Fahrenheit. We had a substantial challenge just keeping the system running.
Another problem was that the drives could record for about an hour, and then it would take us an hour to copy the data from the drives to a host computer, which processes RAW data into image data. Then it took another hour to low-level format the drives. If we had just erased them, due to fragmentation, the drives would no longer have the performance to sustain this high data rate. This was all extremely time-consuming: at minimum, a two-hour recycle process.
We actually developed a solid-state array a couple of years ago trying to solve this. It worked, but could only record for 20 minutes, and cost over $6000/drive! That was never deployed in any practical sense.
We have recently been able to add newer solidstate drives that are affordable, have performance to deliver the throughput we need, and can record for 46 minutes. We can unload the data in 23 minutes, then just empty the trash and be ready to go. Our two-hour recycle period is now down to 23 minutes.
The solid-state drives operate in temperatures as low as zero degrees Fahrenheit, and as warm as 140 degrees Fahrenheit. They are impervious to shock and vibration, and can deal with the kind of harsh environments that had been a problem for hard drives.
TAKING THE 'SUMER' OUT
Some people have a problem that our cameras are based on essentially a prosumer camera system, but that's misleading. It is understandable that camera manufacturers need to differentiate their product lines, but cameras like the HPX300 and the Sony EX-3 are blurring those lines. We also found that, when we opened these cameras up, many of the components were the same whether the camera cost $5000, $15,000 or $50,000.
Once you take those components and add genlock, come straight off the sensor with uncompressed 10-bit 4:4:4 RAW data, user-definable LUTs and so on - you've pretty well taken "sumer" out of the picture, and moved into "pro" territory.
We are also seeing a dramatic difference in the imagery coming out of this camera compared to many others in the market. The dynamic range has increased to about 10 1/2 stops latitude. The resolution has obviously has improved, and the low light capability seems to be improved. There are just so many things about it are far different from the stock camera.
3DVX: THE NEXT GENERATION
The 3DVX has proven to be a real workhorse for us. It's got a lot of what people are looking for in a 3D camera. It's not like a big crazy rig - it's more like just a camera. You can shoulder mount it. You look through a binocular view finder. It's ergonomic. It has a sensible kind of interface to it.
We have incrementally added lots of great functionality to the 3DVX since its introduction, with version numbers to reflect the hardware changes - almost like software version numbers. We had previously gotten to 3DVX 3.5, but have recently added some major new features. We've also added a new, black anodized case, so for now, we're calling this new version of the camera the 3DVX Black.
One of those major new features is something we call "optical axial offset."
Zero parallax is sometimes called "convergence," when the views from two lenses meet. One way to set the distance for zero parallax is to angle the cameras. You point them in toward each other, the centers of the lens views - their optical axes - cross somewhere in space, and that's your point of zero parallax.
The problem is that this makes for a mechanically relatively complicated camera. It also introduces optical distortion into the image in the form of keystoning, and other issues that make it not the most desirable approach.
We've approached it from another direction.
The DVX100 and many similar cameras have an optical image stabilizer. It uses an accelerometer, similar to the one in an iPhone, which can detect the attitude of the camera. As the camera moves around, a magnetically driven lens element attempts to counteract that motion.
We don't like to use that. If we want the camera to be stable, we hire a Steadicam operator. We have disabled the stabilizer - anyone can do it; it's just a software switch in the menu - but we have used the technology to our advantage.
We have actually taken over the drive mechanism, and added four controls that allow us to control the horizontal and vertical position of the light falling on the sensor. Just like lens shift on a projector allows you to move the image around the wall without moving the projector, and without introducing keystone distortion - we do exactly the same thing in our camera. You have four dials to turn to make these adjustments. The optical changes update in the viewfinder in real time, which makes it very easy to use.
Before we added optical axial offset, the camera was realistically operating at its best only at full wide angle. Any slight geometric differences in the images from the two lenses would be magnified as you zoomed in. This was impossible to adjust.
Now, as you zoom in, you can remove any sort of vertical mismatch in the left and right images. You can also achieve zero parallax manipulation without angling the cameras and introducing optical distortions. The 3DVX is the only digital stereoscopic camera in the world that can do that.
Whereas the 3DVX Black is a side-by-side configuration, we now also have a beam splitter camera system that we used on "Call of the Wild." It is based on modified HVX200 cameras that are mounted perpendicular to one another using a one-way mirror, in a fashion that you may seen in other camera systems that have been around for quite a while.
There has been no image quality modification to the HVX cameras we use, although they have been modified for genlock, and had our axial offset system added to them. We call the camera system simply the BX1, one of the smallest and lightest self-contained 3D beam splitter cameras available.
We are currently developing our next generation system, called the BX2. It is primarily being designed to accommodate larger, higher- resolution cameras, including the RED One, the Arri D21, the Phantom, and the Sony EX3.
Pedro Guimaraes, with the 3DVX. Photo courtesy Stereoscope Studios Shannon Benna.
We have just begun shooting a project for the Boston Science Museum and the Museum of Health in Houston, Texas called "Planet You," which will be on the museum circuit soon. It's part of the wide range of live-action 3D projects we have worked on since 1999, many of them corporate and industrial productions for hi-tech companies, pharmaceutical companies, retailers, amusement park rides, and even material for some rock concerts.
We have also just opened an office in Burbank. 21st Century 3D is now the only stereoscopic production company with offices in both Los Angeles and New York. This is helping us to give clients on both sides of the country, and in many different areas within the production world, the best attention, service and creativity we can.
All of this has led to interesting evolutions of our systems. I'm excited by where we're going, both at 21st Century 3D, and the entire 3D industry in general.
Find more great Creative COW Magazine articles by signing up for the complimentary Creative COW Magazine.