The use of ``postage-stamp'' animations with ``fbanim'' can provide useful visualization capabilities. Unfortunately, if there are enough frames to present a smooth sense of motion, the individual images are so small that details are lost. If the images are made larger, the motion is no longer smooth.
Videotape offers the ability to maintain a moderate image size (usually at least 640x480 pixels) with a passable time resolution (25-30 frames per second).
There are several things which must be determined when preparing to make frames for videotape. These are:
The television and video industry has not settled upon a world-wide standard for the encoding of video pictures. Some of the basic elements are common to all the standards Each frame of video is comprised of two ``fields.'' A field consists of either the even numbered scanlines or the odd numbered scanlines from the frame. To display the frame, first the field made up of the odd numbered scanlines is displayed. When this is completed, the field containing the even numbered scanlines is displayed.
The NTSC video encoding system (used in the United States and Japan) displays approximately 30 frames per second. This means that 60 fields are displayed per second. The flicker that would ordinarily be perceived at 30 frames per second is substantially reduced. The PAL (Western Europe and Australia) and SECAM (France & former Soviet Union variant) encoding formats display 25 frames per second or 50 fields per second.
Resolution of Video Recording Formats =========================== Format VLines --------------------------- VHS (1/2 inch) 240 3/4 inch 260 3/4 inch SP 380 SVHS (1/2 inch) 400 Hi8 400 BetaCam SP 400 D2 Digital Video 440There are other differences between the various formats listed. The table is roughly ordered from lowest to highest image quality. If possible, avoid using formats closer to the top of the table for making original recordings. Choose the best format available to you for making original recordings. It is usually easy to duplicate an original to a less capable format for distribution. Plan on making all duplicates from original recordings if possible. Second and third generation copies made with the analog recording formats will show degradation in color (bleeding and noise), image stability (straight lines become wavy), and resolution. If the distribution copies of your animation must be provided on VHS, keep in mind that fine details in the scene may be lost when the transfer is made to this format.
In order to know what size images need to be computed it is necessary to know something about the hardware that will be used to convert the images to a video signal. The video encoding system found in the Silicon Graphics Iris (tm) family of workstations encodes a 640x480 pixel image. The Abekas A60 Digital Video Disk system encodes an image that is 720x486 pixels in size. Other video encoding hardware may use other resolutions. You should determine the image size encoded by your hardware.
Frames should actually be computed at twice the resolution that will be used for the encoding (e.g. 1280x960 instead of 640x480) The ``pixhalve'' program is used to reduce frames to the appropriate size for video encoding. The filter kernel used by pixhalve reduces sampling artifacts such as the pixel staircase effect on diagonal lines (a.k.a. ``jaggies'') in the final image. It also attempts to ``spread'' single pixel details slightly (such as specular glints of light off surfaces). This slight spreading helps to compensate for the fact that current video encoding and recording techniques have difficulty with such fine details.
The ``-J1'' option should also be given to rt. This turns on the ``ray jittering'' feature of rt. Ordinarily, rt traces rays through the center of each pixel. This can lead to visual artifacts resulting from the regular sampling grid. When ray jittering is enabled, rt picks a different location inside each pixel at which it traces the ray. This reduces the regularity of the sampling grid, and hence any artifacts that might result.
The rt program fits a viewing cube with 1:1 aspect ratio in model coordinates to the dimensions of the image being created. This means that whenever a non-square image is created (such as a 4:3 aspect ratio picture for use in video recording) the image is distorted. This is most noticeable when computing images of involving circular objects such the sphere, rcc (right circular cylinder) or the torus. This becomes a problem when preparing images for visual analysis and animation. The ``-V'' option to rt provides the solution to the problem. The following shell script demonstrates the use of the this option.
Video has a very limited capacity for color. It also does not have equal capacity for storing the primary colors red, green, and blue. This usually comes as a great disappointment to people accustomed to looking at computer images in rich detail on workstation monitors. Whenever possible, the colors for objects in a video scene should be chosen to match the abilities of the video system being used.
The first rule of thumb is: if an image or scene is going to be recorded on video, the colors should be chosen by viewing the scene through the video encoder, not on the workstation monitor. A good approach is to render several images from your animation sequence, and record long runs of them in the same manner that will be used for the final animation frames. View this test recording to see how colors will appear in the final result.
Blue objects in a scene are most likely to have visual noise to their appearance. A large smooth blue surface is likely to look ``grainy'' in the final result. Small blue details will get lost in the noise.
Red objects are most likely to contribute to smearing or color-bleeding in the final result. Where possible, avoid placing bright or predominantly red objects next to areas of important detail.
One other consideration worth mentioning is that no color in the scene should have more than a 75 percent level of saturation. While this is rarely a problem with images generated by rt, it is a frequent ``beginner's mistake'' when preparing title frames or other hand-painted images. The ``fbcolor'' program makes the viewing and selection of specific colors easy. It also reports the saturation level of the colors it displays as a number from 0 to 255. You should avoid using any color which fbcolor reports as having a saturation over 191. If you should notice that an object image created by rt lacks depth or surface detail (such as a lighted cylinder which shows no shading along the curved surface), check to see what color it was given in the model. Reducing the saturation of the color given the object should help improve its appearance.
There is a tradeoff to be made between the number of frames computed (and hence the computational requirements) for a sequence of a given duration and the quality of the motion in the result. Larger numbers of frames (and more CPU time expended in their creation) result in smoother motion. Fewer frames per second result in more ``jerky'' movement of the camera and objects.
The smoothest motion is obtained when an image is computed for each field of the final animation. Under NTSC this produces the appearance of 60 separate images per second. This is just above the threshold at which most individuals perceive flicker.
The scanlines for each field must be extracted from the images. The two fields which form a video frame are combined together to create the video frames using the pixfields program. These are then encoded to video and recorded. Besides the computational requirements, the only drawback is that sequences created in this manner do not look as pleasing when used with ``freeze frame'' features common on videotape machines.
The most common compromise is to compute one image for each video frame of the final result. There is no requirement for field compositing, and the image obtained from a ``freeze frame'' of the videotape machine is pleasing. Under NTSC this produces the effect of 30 frames per second.
If each frame is shown for two or more successive frame times, the perception of motion begins to suffer. For any finished product it is advisable to use each image for three or fewer frame-times.
It can take hours, days, or weeks of time for rt to create all of the images in an animation sequence. As a result, there is a greater probability that rt will be interrupted before it can finish the task. If 200 images of a 300 image animation were completed before rt was interrupted, it would be a waste to re-create those images when rt is restarted. Therefore rt was given some heuristics to allow it to avoid duplicating results and wasting CPU time.
Whenever the rt program is asked to create an image, it first looks to see if the image already exists. If the image file exists and is read-only, the image is assumed to be completed and rt moves on to the next image. If the image file exists and is writable, then rt looks for black pixels (R,G,B color 0,0,0) in the image. Since rt never creates image pixels with this exact color, any such pixels it finds in the image are recomputed. If the file is smaller than the final image should be, rt computes only the missing portion to complete the file.
The amount of time required to create the frames of a particular animation sequence can be reduced by employing more than one machine to perform the task. This can be done by giving multiple machines portions of the sequence to create. If the sequence is divided up into frames, each machine can work on a set of frames. Alternatively, the program ``remrt'' can be used in place of the ``rt'' program. Remrt distributes the computational load for each frame across a large number of machines. Each machine is assigned a portion of the image to prepare. The result is assembled on a single machine. For a more complete discussion of remrt see [Muuss90a].
Animation frames can require a great deal of disk space. If the frames for a 20 second animation at 30 frames per second were computed at a resolution of 1440x972 pixels per image, the result would be approximately 2.4 gigabytes of storage. Fortunately, it is possible to reduce this requirement substantially. The rt program can be given a shell command to execute from within the animation script. For example, the file ``ell.proto'' could have looked like this:
% cat framedone.sh #!/bin/sh # Script to be run by rt whenever a frame of an animation is completed. # Should be invoked from rt animation script as follows: # # viewsize 200; # eye_pt 20. 0.0 83.25; # orientation 0.0 0.0 .9238 .3826; # start 3; # end; # ! framedone.sh file.pix 3; # if [ $# -ne "2" ] ; then echo "Usage: $0 basename.pix frame_number" ; exit fi echo $0 $* if [ "$2" -eq "0" ] ; then FILE=$1 else FILE=$1.$2 fi pixhalve -w1440 $FILE > $FILE.sm chmod -w $FILE.sm compress $FILE.sm chmod +w $FILE mv $FILE.sm.Z $FILEThe second ``if'' statment in this script is necessary because rt does not append a frame number to the filename of the first image in an animation sequence. This preserves compatibility with scripts intended for creating a single frame.