SDG here jumping with joy at the thought of seeing two early Pixar classics in theaters, back-to-back, in 3-D!
“You… are… a… TOY!”
Toy Story and Toy Story 2 are back in theaters in a double feature, and they’ve been converted to Disney Digital 3-D. This is in anticipation of next year’s debut of Toy Story 3 in 3-D.
Converting a computer-animated film like the Toy Story films to 3-D is an entirely different proposition from doing a 3-D conversion on a movie like, say, Tim Burton’s The Nightmare Before Christmas (which will also be back in theaters in 3-D in a couple of weeks or so).
The Nightmare Before Christmas is stop-motion animation, which means it was filmed using real 3-D objects in real space. Had they wanted to make it 3-D originally, it would have been comparatively easy to film it in 3-D the same way you would a live-action movie, by using two cameras together, one for the left-eye perspective and one for the right-eye perspective.
That’s how 3-D works: by presenting two slightly different perspectives on the same action to each of your two eyes, which your brain compiles into a single 3-D perception of spatial relations. That’s the way you perceive actual 3-D space too: Your left eye and your right eye have slightly different perspectives on the world, and your brain does the math of mapping how close or far objects are depending on how different the perceptions are from each eye. The further away objects are, the less difference it makes to your left eye versus your right eye; the closer objects are, the more different they appear and the more their position shifts relative to one eye versus the other.
For example, your left-eye view of your nose is completely different from your right-eye view of your nose, because your nose is really, really close to your eyes. Looking at these words on the computer screen, if you close first one eye and then another, you’ll see the words jump slightly to the left or the right — but not as much as your nose, which jumps completely from one side of your visual field to the other side. Then if you look out a window at objects that are far away, they shift even less.
Of course, whenever you have one eye closed, you’re seeing a 2-D view on the world, just like a photograph or a movie, although your brain still understands space well enough to work with a 2-D picture — though not as well as if you can use both eyes.
Among other things, even with one eye, your brain makes judgments about space not only based on the same sort of judgments you make looking at a photograph, but also as your head moves your brain gathers additional information about what things look like from different perspectives, and uses that information to make better judgments about distance.
Ever seen a cat bob its head up and down before making a jump? Same thing — it’s gathering more information to make the best possible estimation of the distance. It’s almost like having four eyes instead of just two. (I think maybe some athletes, like tennis players, might get a similar benefit by swaying back and forth, though that’s probably mostly about being ready to move in any direction.)
“What’s this?!”
Anyway, getting back to The Nightmare Before Christmas, although it would have been comparatively easy to film in 3-D originally, in fact like any non–3-D film it was filmed with a single camera point of view, which means that now all the information about those objects in space has been reduced to a 2-D image representing a single point of view — not enough to create a 3-D image.
Creating a 3-D effect in that case means extrapolating (i.e., creating) additional information that doesn’t exist on the film about what those objects would look like from two different points of view, as well as what we would see of objects behind them if we had a slightly different perspective, etc. In other words, you have to cheat and make stuff up. Fortunately, computers are powerful tools and the effect is pretty good, though not as good as filming in 3-D in the first place.
With Toy Story, though, it’s completely different. The great thing about computer animation is that even though the film images were rendered by computers in 2-D, prior to being rendered the films were staged and animated in a virtual 3-D environment — and all that lovely 3-D information still exists on hard drives at Pixar. For example, in a scene in which we see Andy playing with Woody and Buzz, we see them on screen from only one perspective — but the animators originally mapped out where Andy, Woody and Buzz were in relation to one another in virtual 3-D, and the computer files with that information still exist.
In principle, the animators could swing the virtual “camera” 180 degrees around the room and render to show us Andy, Woody and Buzz from the back — or what it would look like from a bird’s eye view over their heads, etc. You could never do that with a 2-D film like The Nightmare Before Christmas — you’d essentially be painting an entirely new image with all-new information.)
To give us 3-D, though, Pixar just have to render two different points of view similar to the original camera angle for a left-eye and a right-eye shot.
Then both images are projected on the screen at the same time, and the images are filtered for the left eye and the right eye using polarization, i.e., controlling how the light waves move for each of the two images and then using polarized 3-D glasses to filter for light traveling in one direction versus the other. (In the old days of 3-D, polarization was linear, e.g., vertical or horizontal, but newer circular polarization, which is left or right, is much better and doesn’t depend on the angle of your head. With the old linear-polarized glasses, if you put two pairs of glasses together and turned one at right angles to the other, you would see almost nothing. That wouldn’t work with circular polarized glasses — but I bet it would if you opposed them face to face.)
None of this, of course, has anything to do with why Toy Story and Toy Story 2 are such classics … for that, you can read my reviews. (Oh, and The Nightmare Before Christmas, not exactly a classic, but quite a fun little film, and fun to revisit around Halloween.)
Next week: The return of the Petrine Fact!
I’ve been wondering if the renderings will be BETTER than initially released. Pixar’s software has evolved since then. Hair and people’s faces have improved dramatically. Just compare Toy Story to Toy Story 2 and you can see the improvement.
So I could see them running the old 3D models thru new and improved rendering engines. It could be MUCH better.
According to Pixar Planet, yes, textures and such have been enhanced, but AFAIK the 3D models have not been altered at all. Certainly there are no new scenes or anything, they’re exactly the same films in that respect.
Something else to consider: According to Watching Apple, back in 1995, one frame of Toy Story took an hour to render. With new computers and software, it took less than a 24th of a second.
“More than 86400 times faster than 14 years ago,” the site says. But if it’s reported correctly, it’s actually twice as fast as that, since each frame includes two separate renderings. So that would be 172,800 times faster!
I got up one morning in a very dark hotel room, and still a little groggy, walked over to the window to look outside. Something didn’t seem right with my sight, but I couldn’t quite figure out what, so I covered my right eye — everything pretty much looked the same. Then I covered the left — complete darkness. Ah, there’s the problem — somehow my right eye had goo’d shut.
Anyway, thanks for the interesting 3D explanation.
From Wikipedia: “In total, the film required 800,000 machine hours and 114,240 frames of animation, with 2–15 hours spent per frame.”
Of course, you can’t exactly scale that linearly from 800,000hrs to 9-18hrs (depending on whether ‘1 frame’ means one or two images.) Changing things like the hair model and whatnot probably introduce some extra overhead.
Here’s a question, what’s the resolution of the original rendering? Is the new resolution the same? Anybody know this?
Could be, but the source for the previous stats is a Wall Street Journal Online article interviewing John Lasseter, which says, “Where the original ‘Toy Story’ required an hour per frame to create, Mr. Lasseter said, rendering the new 3-D version took less than 1/24th of a second per frame.” If the paraphrase is accurate and Lasseter really said that, it’s probably at least close to the truth.
In any case, the Wikipedia article above seems to be speaking about the original 1995 rendering of the film, not the 3-D re-render. If, contra Lasseter, 2-15 hours is more accurate than one hour for the original render, and the 1/24th of a second per frame rate reported for the re-render is also accurate, then the rate of improvement is vastly enhanced yet again.
But now I’m wondering if the WSJ journalist might have made a mistake in reporting the speed of the re-render. I’m sure the re-render was indeed accomplished in a fraction of the original time. But 24 frames per second just happens to be the projection rate for most films today. The WSJ piece thus implies that the 3-D re-render was crunched in real time, which seems staggering.
But the article doesn’t make that connection to real time or projection rate, which seems an odd point to miss. Is it possible that Lasseter said something about the frames and 1/24th of a second — referring to projection time — and the journalist mistook this for a reference to the rendering time?
Maybe not. For all I know, it’s possible that today’s computers can actually render a movie like Toy Story in real time, even in 3-D. But it’s also possible, AFAIK, that the reporter made a mistake.
If you happen to be a geek, the number 86, 400 might look familiar. It is the number of seconds in a day = 24 hrs/day x 60 min/ hr x 60 sec/ min. This data looks bizzare. What the original data is saying is that if it had taken a day to process one frame in 1995, it would take only 1 second, today. If it took 1 hour (3600 sec = 60 x 60), it would take 3600/86,400 = 1/24 of a second. In other words, they shaved a days worth of seconds from the processing. I don’t doubt they speeded up the processing time, but I highly doubt they did it by exactly this much.
In 1995, computers had a maximum clock speed of about 30 MHz; today, about 3 GHz or an increase of about 100 times. If they were using Macs, this speed change, alone would only make the process 100 times faster. I suspect they were using Silicon Graphic workstations (I used the same model for my doctoral work). That speed is measured in flops – floating point operations. Could there have been this much speed up? Actually, yes, depending on what set-up they are using compared to the original set-up, whether it is vectorized, etc.
In fact, for a mere $7000 Silicon Graphics will sell you a desktop supercomputer. That is how much faster processing has gotten due to multi-core processors.
The Chicken
Taking the kids to see Toy Story 1 & 2 in 3-D today. Can’t wait, even though I’ve seen each one a billion times. 🙂
Then both images are projected on the screen at the same time
Disney Digital 3-D
“…144 frames per second, six times as fast as a normal movie. Every 1/24 of a second (the projection frame rate for normal 2D movies on film) the two scene views called “right eye” and “left eye” are each shown 3 times (6 flashes of image on the screen matching the 6 times higher projection rate)… In front of the projector lens sits the Z-Screen, an electronic device developed by RealD. It inserts a polarizing screen that matches the polarization of either the right lens or left lens of the glasses worn by the audience. When the left-eye-matching Z-Screen is in place, the viewer’s right eye sees nothing at all (or almost nothing) while the left eye sees a normal looking frame. For the next frame of the movie, the Z-Screen swaps the polarizing screen to match the right eye lens in the glasses worn by the audience. Now the audience sees nothing (or nearly nothing) with the left eye and a normal but slightly shifted version of the frame in the right eye.”
Ah, thanks for the additional info, Terry.
Yes, of course the projector has to alternate right and left views (how else would the polarization work?); by “at the same time” I didn’t mean “in the same actual frame of film or otherwise in the same instant of projection,” although I did sort of assume that the same instant of film action from the 3-D models was rendered twice for the the right and left views, though I see now that they can stagger that as well, which makes perfect sense. And I had no idea the frame rate was so dramatically increased for Disney 3-D, nor did I know that Disney 3-D was projected exclusively in digital, never in film.
Incidentally, most people don’t realize that for normal film movies (i.e., movies projected from actual film, as opposed to digital projection) the screen is actually dark between frames — that there’s nothing on the screen as often as there’s something. A shutter cuts off the projector between frames as the film advances from one frame to the next, so that each frame is projected as a still image, followed by another still image, with a split second of darkness between them. Otherwise, you’d see the blurry motion of the film whipping from one frame to the next.
In fact, for most movies the shutter speed is actually faster than the frame rate, so that each individual frame is projected, cut off and reprojected more than once before the frame advances. That assumes a normal 24-frame-per-second film; I have no idea if a similar principle applies to digital projectors, as with Disney 3-D.
Here is an interesting question: films are shown at 1/24 of a second per frame. How accurate is this? How much variation from individual to individual is there? I can’t imagine that this is a fixed rate for all human beings. Is it a greater lowest bound? Is it possible to “tune” the rate for individuals (perhaps a speed knob on a dvd player? What effect would that have, physiologically?
The Chicken