Captain 3D Presents...Volume, Lenses and other ideas |
In Real life we can see great stereoscopic depth from just 6" in front of our nose all the way out to infinity. Converging this close in a movie theater would require stereo separations that would be as wide as the 40ft screen itself projecting a 1920 pixel wide image. For most people this is extremely uncomfortable if not impossible to view. From my own experience I have found most people can comfortably view a stereo movie image with a total of 50 pixels (2.5%) of separation split either side of the screen. This is about twenty times less depth than real life which brings up some interesting depth challenges. The picture below represents about a 62 pixel (3.2%) depth bracket at 1920 but it gives you an idea. |

Oscar from Chile - 1993
Ideally we would have full freedom to create stereo depth that represents real life as the base line from which we could make creative distortions for the purpose of our stories, but we have to distort depth just to get it into the comfort zone of the 3D theater. We almost never distort an image in the X or Y dimensions for 2D cinema so why would we want to distort space in Z? The golf balls below have been spatially distorted (in a none scientific way). The top row of golf balls represents distortion only along the Z axis achieved by eye balling different interaxial settings. This compression has been very common in recent live action 3D movies. Number one is grossly flattened while number four is supposed to represent 100% roundness. View the image from about arms reach. The bottom row represents the same level of Z distortion but also applied to Y so we can make a visual comparison of the 3D and 2D distortions. What level of distortion is acceptable in 2D and 3D? |
Is anything less than 100% round acceptable in 2D? To my eye you might get away with the distortion of lower ball number three in a 2D movie but we would still consider it an error. 3D is a bit more forgiving and maybe upper ball number 2 is still an acceptable depth representation. In my experience a 40% depth compression (60% volume) can be mixed with other shots of greater volume without things looking too strange. Below 60% volume and characters start to look off model. When stereo space is flatter than 60% volume, the set starts to look like it is in a spatial warp field. When characters or cameras turn in this warped space or try to make eye contact with another character things look very strange. |
On the first frame, of the movie below, our character looks very wide from left arm to right arm but very flat front to back. By the time the camera has swung around by 90º our character now looks deep from chest to back but flat from arm to arm. That is a Z depth warp field. |
Pump-Action in a Z depth warp field
The Theater space itself is another form of distortion to think about. The middle seat in a theater creates a field of view to the screen that is very close to our 24mm lens (54.8º on our CG camera which is supposed to match super 35 I think). So there is no surprise to find that shots created with a 24mm lens fit the theater space much better than a 50mm which has a capturing field of view that will be translated into a 24mm viewing field of view by the audience member looking at the screen. The top row of the next diagram represents the theater field of view as experienced by the audience. The bottom row represents the lens field of view used to capture the scene. If the two do not match something is going to get squashed. Very wide lenses can also stretch a scene. |
Below are some movies comparing different lenses used on the same
shot.An optical stereo window is used on the right hand side.The first
movie is really two movies in one. The top half includes the characters
and background. The lower half only has the characters. I have tried
to set stereo in a way that I think keeps the character volume consistently
and acceptably round through a range of focal lengths. |
Scrub through the movie until you find a 'lens look' that you like. When happy, stop scrubbing then scroll back up the web page to see your choice with background and number data. Sit as far back as possible. Even though I know the movie data I am surprised at myself picking lenses that turn out to have way too much background separation for a 3D theater. The longer lenses just look better compositionally (Or is it that we are just used to that look from years of 2D imaging?) but will cause discomfort or require crushing the volume. In this movie the depth bracket radically expands with the longer lenses as I have cranked up the IA to try and preserve character roundness.
The next movie has a fixed depth bracket of approximately 50 pixels split either side of the screen. This is the amount I consider comfortable for most viewers. As the depth bracket is fixed, the nearest foreground and farthest background will remain pinned in depth. As the focal length increases, the narrower field of view actually needs greater depth to look volumetrically correct but our depth bracket is fixed. The result is that the greater, captured depth is crushed into the fixed depth budget. It is like trying to close the zipper on a badly packed suitcase. |
At the wider end of the lens range the character volume, relationship with each other and background depth all work well together. As the lens length increases the shot remains comfortable due to the fixed depth bracket but the volume quickly collapses and the characters quickly turn into cardboard cut outs. The spatial relationship between the characters also enters the warp field as the Z depth gets crushed but X and Y dimension remain true. The character on the right almost looks like he is moving in for a bite as you scrub to the right. A trick I use often can be tried here. Scrub the movie to the last frame so you are framing through the 135mm lens. Keep looking through your anaglyph glasses but close your left eye. Viewing the frame with one eye removes all stereo data and allows your brain to imagine the true depth of the scene. When you think you have imagined the depth correctly, open both eyes. You should be immediately confronted with the distortion of this lens and stereo combination. It looks horribly flat. Cardboard cutout characters are too close then there is a big gap in space before reaching the cardboard cut out background. Now try it again with the wide lens. The fit is much better. For this technique to work you must close one eye. You can not use two eyes and look at a mono image as this gives your brain stereo data that the image is flat. Closing one eye lets your brain imagine it is a real spatial scene but you just have an eye closed. This technique can be used on set or in dailies. When you are unsure if a shot looks good just close one eye and let your brain imagine the depth. Then open both to see if your shot matches the spatial world model already inside your head.
This last movie clamps the far separation to 24 pixels (1.25%) and keeps our center characters eyes on the screen. I have left the screen right stereo window unchanged so you can see how much this method collapses the nearest character into the scene as the lens gets longer. |
|
This method produces the flattest shots of all. In all three movies I start to have problems with either the character volume, comfort or general depth integrity of the screen once the lens goes past 35mm. It is no surprise then that in CG we use 'Multi Rigs' or multiple stereo settings comped together when lenses get longer than 35mm.
In conclusion you have a choice on every 3D scene. Lens palette, character & scene volume and viewing comfort. Presuming you always want viewing comfort you have to find a balance between focal length and scene volume. Once you go past 40mm or 50mm on the focal length you have given up on finding balance for most shots in a movie. Phil McNally |

The Triangle of Choice