On July 17 and 18, 1967, a conference on computer animation for motion pictures was held in Newton, Massachusetts. This conference was similar to one held two years previously at Bell Telephone Laboratories, and was organized by the Commission on Engineering Education and the Users of Automatic Information Display Equipment (UAIDE). Attendees included those with a need for films (primarily educators); film makers (specialists in composition, timing, use of effects, etc., in motion pictures); and computer display specialists. Since the proceedings of this conference were not published and are probably not available elsewhere, the following summary of the papers, tutorials, and films presented has been included in this report.
Terrible color and texture, and difficulty of instruction were named as disadvantages of the computer as an animation tool. Lower cost was judged a debatable advantage. The recommendation was made that this technique be limited to those things that can only be said by use of a computer. For example, a scaled universe might be created for students of quantum mechanics, astrophysics, or relativity, and they might be shown experiences in that universe in order to develop pedagogical intuition.
A quantum mechanics film was shown, first in black and white, then in blue and white, to illustrate how even a small amount of color can create a more benign learning atmosphere. Technical problems encountered in making this film included the need for a stable camera with pin registration and the need for a computer display fade-out method.
The point was made that in display programming the simplest approach may be the most costly. For example, in the quantum mechanics film, y = e-x was plotted: it was estimated that the two-instruction approach would have required 20 hours of computer time per minute of film time. The 20 hour figure was reduced to 12 minutes by programming
Yj+1 = e-ε Yj where ε = xj / j
which requires that the exponential routine be entered only once. This generation of increments is consistent with random positioning displays but requires care in keeping track of cumulative errors.
A two-dimensional projection of a three-dimensional object is not unique. This is why image tumbling or rotating is useful.
He showed a fluid dynamics problem film generated with a CDC 6600 and SC 4020. He reported that particle representation proved less costly than contour representation of fluid flow, quoting 8 seconds of 6600 time per film frame for particle representation. Some Moire interference patterns appeared in the images as a result of the 4020 mesh size of 1024 × 1024.
He showed three dimensional movies, viewed with stereoscopic polarized glasses. His conclusion was that 3D movies can easily be produced using subroutines.
He discussed the economic and technical considerations of computer animation. His remarks were based on experience producing a hydrodynamic film under a government grant.
In order to obtain reasonable quality of text a 20 × 30 point matrix had been used for character generation. An image element used extensively in the film was a vector arrow; accordingly a subroutine was developed for arrow construction.
In order to overcome image jitter a professional Bell & Howell stop action camera with locking pin registration was obtained and jury rigged to a SC 4020.
The maximum line segment that can be programmed for the SC4020 is 64 raster points. Breaks appeared between segments when constructing a longer line than this. Therefore each line was retraced, backwards, in order to cover up these interruptions.
Exposure settings were obtained by measuring the light output of the 4020 and adjustments were made manually.
The price of the first film produced was approximately $2500 per film minute. The price for another film on a similar subject (allowing the same subroutines to be used} would be quoted at approximately $600 per film minute. These figures compare with $1000 to $2000 for hand animation.
Computer execution time was tape limited by the IBM 6450 at 6500 to 80000 lines per computer minute, or 65 to 80 lines per penny. Six tenths of a second per frame (l - 2 cents) was required to advance the film.
The primary factor in the high first-time cost was computer program debugging. With the use of subroutine packages this could be reduced to less than one fourth of production cost. Another technique for reducing this cost element is to produce only every tenth frame on a debug run.
Economic pitfalls cited were numerical analysis problems; poorly detailed storyboard; and last minute changes to the storyboard. Kaye's initial experience was 50% of film cost for programming, but this is now down to about 30% because old routines are being put to use. Hopes are to get the figure down to under 20%.
He spoke on Self-Explanatory Visualization, a teaching technique based on pure pictures (no titles, no text) which provides a medium for the teacher remarks-to-student understanding expansion analogous to the FORTRAN-to-machine language expansion.
Also reported was the development of a movie making language for computer animated line drawing and perspectives. The work of Huggins and Knowlton in animation languages was acknowledged.
He stressed the importance of selective movement in a picture, achievable by grouping image items in stacks. For the reduction of animation costs it was recommended that a mechanical plotter be used during debugging, rather than the microfilm recorder.
He spoke on techniques for producing color movies. These included the Sylvania two-color tube; a white phosphor CRT; (e.g. P4) with a filter; 3 CRT's and dichroic mirrors; a shadow-mask color CRT; and as practiced at LRL, the generation of sequenced images for each color on black and white film, followed by the electro-optical conversion (using sequenced filters) to color film. A savings of 50% was quoted using the DD-80 rather than a SC 4020. The DD-80 operations take from 1/8 to 1/3 the 4020's time.
He gave a tutorial on film making. At 24 frames per second, 2-frame animation (same picture twice) is satisfactory for certain types of motion and reduces preparation cost. The quality of 3-frame animation is not considered acceptable under any circumstances. Distinctions were made between raw footage and finished film, and between animation and live action. The use of storyboards was described, as was the use of clapsticks to synchronize soundtrack and film. Work with a moviola editor was described as 98% bookkeeping, 2% art.
The Bell and Howell camera with pin registration allows backing up for double exposures on the SC 4020, with good results.
The Bell and Howell standard movement is used in animation cameras. A less expensive camera with similar features, which like the 4020 camera is designed for animation, is the 18-frame per second $3695 Giannini.
He spoke on How to Get Started. Reference was made by John Carr to the fact that movies were made on the Whirlwind in 1951. Today a typical configuration consists of the following equipment:
An alternate approach is found in the Japanese Hitachi system which uses analog information. Available equipment costs are approximately as follows:
Equipment Monthly Purchase Remarks
$ $
DEC 338 60K+camera Designed for Direct Viewing
IBM 2250 1500-2000 Designed for Direct Viewing
+ camera
SC 4020 800 130,000 With automax camera
CDC 280** 5800 170,000 With Giannini camera
BL 120 4500 160,000 With automax camera
Calcomp 80,000 off-line Incremental plotter
835CRT 60,000 on-line Incremental plotter
* Educational Discount Price
** Old DD-80
An alternative to buying the equipment is to take one's magnetic tape to someone else's machine. But at $0.60 per frame, or $864 per film minute (6 minutes of 4020 time) one would pay for a 4020 purchase in one day. The following are sources of 4020 time rental:
Area Source Rate Cameras
---- ------ ---- -------
Boston AVCO $0.50/f to .05/f over 10Kf 35 mm 4020 camera
New York Brooklyn Poly $200/hr + $1.44 per min. 35 mm B&H
of film for processing l6 mm non registering
Boston J. Kaye $135/hr + $30 setup 35 mm B&H
El Segundo Alpha Graphics $100 - 150/hr 35 mm 4020
Philadelphia GE Va. Forge $0. 60/$ 35 mm 4020
Boston Lincoln Labs ?
Washington DTMB ?
The 4020 FORTRAN SCORS package is available for a number of machines from the UAIDE librarian at Stromberg Carlson, San Diego. The 4020's come in two sizes; 1024 or 4026 character buffer. SCORS was not designed for animation, but the New Brooklyn Polytech Package (NBPP) is available from Zajac. This NBPP will include vector algebra, save-routines for backgrounds, line drawing routines, automatic windowing. Also BEFLIX is available from Knowlton at Murray Hill, but the Bell Operating System is required to run BEFLIX. In order to debug before running on the 4020, IBM has a simulation program, written for the 7094-1401 which produces sample frames on a 1403 printer.
Comparing the DD-80 with the SC 4020, the DD-80 is faster (approximately 20% of the 4020's time).
He offered comments on professional quality considerations. At Lincoln Lab he has used an old slap action camera mounted on a tilted forward 4020 CRT.
A RackOver mechanism or Acadamy Frame is used to collimate the camera.
For proper exposure photocell sensors are used and CRT intensity control is adjusted as required by film response. Beam size must be right for use with the extruded beam character stencil. A diagnostic tape with 50 images is used in test runs.
He discussed non-computer animation, in particular experimental techniques.
In classical hand animation a series of figures are drawn and photographed. Grumbacher celluloid is frequently used for backgrounds, separate drawing of head and feet, etc. This art is on its last legs.
Photo-animation involves zooming in and around a static photograph.
Experimental techniques include scratching, sand papering, perforating film, painting cellulose ink (Pellican T-type) on film.
Other approaches include the use of puppets and clay figures. Cheapie hand animation can now be done for $l/foot. (?)
They replied to Michaels' comparison of SC4020 and DD-80. The SC 4060 will have a 3 × 4 frame for movie making; 12 dot sizes for facsimile style tone pictures; and two or more line weights.
Character sizes will be expandable to 64 raster points on the 3072 × 4096 (3 × 4) array. This is four times that of the 4020 so that characters should show up OK on film.
Continuous lines will be available rather than segments. Line drawing speeds will be 1 msec corner to corner for light lines, 4 msec heavy lines.
The 4060 will be adaptable to economic on-line use, with a buffer. An optional pin registered camera will be available at $8K (compare with $4K Giannini).
He cited the five instructions ADVANCE FRAME, SCALE, POINT, LINE, CURVE as an adequate set for line drawing, therefor for animation.
The concept of treating each picture element as a subroutine was described. In five FORTRAN instructions the following steps can be accomplished.
l. calculate picture coordinate data 2. scale data 3. reduce coordinate to 4020 instructions 4. draw
He described BEFLIX a software package with which a picture area mosaic matrix of 252 columns and 184 rows can be painted by programmable bugs.
Each computer word contains 3 bits of gray scale information for 12 tiles on ths mosaic. Using BUG language the programmer talks to 26 bugs, each a group of 8 registers containing:
1. The pointer to the tile the bug sits on, 2. Width of the surface in machine words, 3. X(bug) 4 Y(bug) 5. Indexing origin address 6. X max. 7. Location of bug 8. Y max
The reason given for the redundancy is to have information immediately available, without calculation time.
He, spoke on data structures. In a FORTRAN example, record forms (data structures) are single variables or arrays. A picture, however, usually maps more conveniently onto a chain. If a chain is used, means for filling blanks with data or pointers, and means for following chains are needed.
The kinds of things represented in a program, or in a data structure which is later interpreted by a program, are:
1. structure and connections 2. non-displayed supplementary information.
This non-procedural specification allows conflicts of conditions. Core size may be insufficient. A pointer may take control to a cell iin slow access auxiliary memory. Recommends consideration of L6 for use in building higher level languages.