Vivid 2.0 March 14, 1992 Copyright 1989-1992 by Stephen B. Coy Table of Contents Introduction . . . . . . . . . . . . . . . . . . 2 Changes for Version 2.0. . . . . . . . . . . . . 3 Running Vivid. . . . . . . . . . . . . . . . . . 5 Statistics Display . . . . . . . . . . . . . . . 7 Legal Stuff. . . . . . . . . . . . . . . . . . . 9 Acknowledgements . . . . . . . . . . . . . . . . 11 Input File Format. . . . . . . . . . . . . . . . 13 Preprocessor . . . . . . . . . . . . . . . . 15 Studio . . . . . . . . . . . . . . . . . . . 17 Lights . . . . . . . . . . . . . . . . . . . 24 Surfaces . . . . . . . . . . . . . . . . . . 28 bump mapping . . . . . . . . . . . . . . 31 solid texture. . . . . . . . . . . . . . 33 Mandelbrot . . . . . . . . . . . . . . . 36 Geometric Primatives . . . . . . . . . . . . 37 sphere . . . . . . . . . . . . . . . . . 37 ring . . . . . . . . . . . . . . . . . . 38 polygon. . . . . . . . . . . . . . . . . 39 triangular patch . . . . . . . . . . . . 40 cone . . . . . . . . . . . . . . . . . . 41 Transformations. . . . . . . . . . . . . . . 42 Clipping . . . . . . . . . . . . . . . . . . 44 Support Programs . . . . . . . . . . . . . . . . 46 img2gif.exe. . . . . . . . . . . . . . . . . 46 paste.exe. . . . . . . . . . . . . . . . . . 50 up.exe . . . . . . . . . . . . . . . . . . . 51 down.exe . . . . . . . . . . . . . . . . . . 52 File Formats . . . . . . . . . . . . . . . . . . 53 Bibliography . . . . . . . . . . . . . . . . . . 54 Index. . . . . . . . . . . . . . . . . . . . . . 55 1 Introduction to Vivid 2.0 Surprise! Yes, 2.0 is finally here. I'd like to appologize to those of you who have been waiting for many months for this to come out. I've been busy doing my Master's thesis and, of course, adding "just this one more feature." A few bugs were squashed in the process although I'm sure that new ones have been breeding in the walls just waiting until this release. Even with the bugs I still consider Vivid to be one of the finest ray tracers (I've) ever written. In the archive you received should be this document, the Vivid executable, some sample input files, and tools for processing the image after it has been generated. Vivid.exe has been compiled to run on any MS-DOS based system with a 286 or better and a math coprocessor. Ray tracing can be extremely slow and anything less is really not worth it. If, however, you don't believe me feel free to write and I'll send you a copy compiled for an bare 8086. I hope you don't plan on using your PC in the next month. Back to the 286/287 version. This version also contains a DOS extender allowing Vivid to take advantage of any extended memory you may have. For those of you that still want something better I also have 386 versions available. See the Legal Stuff section for more info. The rest of this document explains the details of how to run Vivid and the tools and how to construct an input file. I recommend that you print out this file and a couple of the sample input files when first learning Vivid. An example is worth a thousand words, or at least the paper it's printed on. 2 (Most of the) Changes for Version 2.0 I learned to spell Mark VandeWettering's name correctly. Sorry about version 1.0 Mark. Inside/outside problems have been resolved eliminating the need to have spheres with negative radii or worry about clockwise vs counter-clockwise polygons. Polygon patches are a lot healthier. I plugged my brain back in. The description of fuzzy spheres given in the 1.0 docs is wrong. But it's better now. When using the depth of field option a new "samples" parameter has been added to allow you to control the number of rays per pixel used. Spherical light sources also allow the "samples" parameter to determine the number of shadow rays shot toward them. There is no longer a limit on the number of lights except for memory and your patience. The limit was 20 but this info never made it into the docs for version 1.0. The limit on number of objects has also been removed although due to the operating system an effective limit of about 2000 objects still remains. The DOS extender version is limited only by memory. The number of objects contained within a bounding box may now be specified. The default, 4, is equal to the prior, fixed value. This can be specified by adding bunching 6 to the studio structure. Changing this value will affect the performance of the automatic bounding box construction. Results are image dependent but I think that 4 is a good default value. A bug in the camera model caused the image to become distorted when the up vector given in the studio structure wasn't perpendicular to the direction of view. It has been fixed. The traditional flat projection model has been joined by not one, not just two, but by three other projection models, spherical, othographic and parallax. See the section on the studio structure for details. The input file format has been changed to eliminate the need for semicolons and equals signs. In order to be compatable with the old file format, they are still accepted but are now treated just as a space would be treated. A quick-render mode has been added to help speed up 3 previews. Simple math operations are now supported by the parser. These operations are supported on numbers and on vectors (rgb triples and xyz coordinates). The following is a list of the operations supported. Vectors operations supported include cross product, addition, subtraction, scaling, and dot product. Numerical operations include multiplication, division, addition, subtraction and exponentiation. Sine, cosine, tangent, arcsine, arccosine, arctangent and square root functions are also supported. Cylinders can now be specified by giving a cone a single radius parameter. Preprocessor is here. include files, defines, undefs and lack of fixed colors which are now replaced with color.vc. Global transformations are now supported. Together with include files this should make scene modeling much easier. Primitives can now be clipped by planes, spheres or cone allowing easier generation of more complex models. The background can have a palette mapped to it rather than just having a single color. A new bunch of command line switches are now supported. Typing vivid on the command line without any arguements will give you a listing of them. The keyword "noise" is now "turbulence" to better follow the literature. 4 Running Vivid Vivid input files use the extension .v and the output files use .img. With the preproccessor in Vivid, include files are also supported. These may have any extension. The syntax for including a file into your input file is: #include filename Include files may have any extension you choose. The following is a list of the file extensions that I regularly use. If we all use the same convention it should make things easier for trading input files. Feel free come up with your own conventions if that suits you better. .v Main input file for an image. Required. .img 24-bit output file. .map palette file for img2gif or background mapping. .vc include file with color definitions. See color.vc. .vo include file with object definition. .vs include file with surface definitions. The .img files produced by Vivid are full 24-bit color. The details of their format are explained later in this document. Since most systems can't display 24-bit graphics I've included a tool to convert the .img files into .gif files which can be displayed using vpic or cshow. This tool is called img2gif. Its workings are also detailed later. In general the procedure for generating an image will look like this: 1) Create an input file, test.v, with your favorite text editor. 2) Run Vivid. C>vivid test 3) Wait until done. This can take from a few minutes to a couple of days depending on the complexity of the image and the speed of your machine. This is a good time to raid the fridge. 4) Convert the result, test.img, into a .gif file. C>img2gif -m -d test The -m flag tells img2gif to use median cut to determine the palette. The -d flags turns on Floyd-Steinberg dithering. 5) View the image. C>vpic test.gif Because of the way I implemented the preprocessor Vivid creates a temporary file on the disk called xyzzy.v. At this time, Vivid does not delete the file after it is done using it. 5 The reason I leave this is that it is sometimes useful to look at when trying to figure out what the preprocessor is doing with your input definition. Feel free to delete it at any time. Vivid supports a number of command line flags. These flags allow you to change the operation of the program without changing the input file. The following command line flags are now supported: -s Run in silent mode, ie no statistics display. -r Resume. -i x y Set image size to x by y pixels. -n No_shadows, same as studio flag. -d # Set maximum recursion depth value -a mode Set antialias mode. Valid modes are: none, quick, corners and adaptive. Since I can't type I've made it so that you only have to give the first letter of the mode for it to work. -p Don't use the preprocessor. Vivid has the ability to resume the generation of an interrupted image. If during image generation you need to stop, hitting ctrl-C or ctrl-BREAK will cause Vivid to stop at the end of the next scan line. This may be up to a few minutes for a slow image. This only works if statistics are enabled (the default). Later, the image generation may be resumed by using the -r flag on the command line. 6 Statistics Surprisingly enough some people are curious about the statistics that Vivid displays while processing. Somehow they actually got the idea that the numbers may contain useful information if only they were documented somewhere. The secret is out. They don't mean a thing. I just put them in there because I like watching them change while I'm waiting for the image to generate. Sorry to break your bubble. OK, fine, I'm joking, happy now? When you run Vivid it displays the current version number and the copyright notice. It then displays the filename of the input files as the preprocessor digests them. Next a running total of lights and primitives is displayed as the parser reads the preprocessed input file. Once the file is completely read in the automagic bounding box code is called. As it generates bounding boxes you see the number of primitives climb. Upon completion of the bounding box generation the scene extent is displayed. This is the minimum and maximum values in each axis for all the primitives. At this time the first scan line is starting to be rendered. When it is done the screen will clear and a full display of current statistics will be displayed. The meanings for each of the numbers will now be described in glorious detail bereft of all attempts at obfuscation. I hope this makes sense. The resolution is the size of the final image in pixels. The line display next to it shows the progress of the ray tracer. For example, if you where generating an image 320 pixels wide and 200 pixels tall the resolution would display 320x200 (unbelievable!) and the line value would go from 0 to 200. In quick mode this number will increment by 6 instead of 1. The next four lines contain information about the number of rays cast. Eye rays are the rays cast from the viewpoint out into the scene. Reflected rays are those generated by a specular surface and refracted rays are those generated by transparent surfaces. The total is just the sum of these rays. If your input file doesn't have any reflective or transparent surfaces these numbers will stay 0. The next section contains information about shadow rays. Shadow rays are the rays shot from the point of intersection toward each light in order to determine if there are any other primitives in the scene blocking the light (casting a shadow). Vivid uses a shadow cache to try and speed up this process. Cache hits signify an instance when the cached object was found to cast a shadow and no ray was cast. The cache percentage is just the number of hits divided by the number of shadow rays. 7 The next number is the average number of rays cast per pixel. This number can give you a fair idea of just how hard the ray tracer is working to generate the image. With no antialiasing, reflection, or refraction this number should be 1.0. Adaptive antialiasing can push this as high as 16 depending on the amount of extra work it does. Reflected and refracted rays will make this even higher. Using the quick mode this number may go to as low as .066 or so. When this number is less than 1 that means that the ray tracer is guessing at some of the pixels instead of actually casting a ray for them. The average queues per ray value has to do with the bounding scheme. The automatic bounding box generator creates a tree-like heirarchy of bounding boxes for the system to test each ray against. If a box is intersected by a ray then the objects in the bounding box (either other bounding boxes or primitives) are entered into the priority queue for further intersection testing. The "bounds checked" number shows how many bounding box intersection tests were performed. The "queue inserts" number shows how many object passed the bounds check and were put into the priority queue. The "queue resets" number displays the total number of times the priority queue was reset. This should be equal to the total number of rays plus the number of shadow rays plus the number of cache hits. Don't ask why, it just will be. The "max queue size" number shows the maximum number of objects in the queue during the rendering. The final number, the max recursion depth, shows the maximum number of levels deep the ray tracer had to go while rendering the image. The number after the slash is the max depth it is allowed to go. 8 Legal Stuff Vivid is Copyright 1989-1992 by Stephen B. Coy All Rights Reserved. You are free to redistribute this package in its entirety. In doing so you may charge no more than a nominal fee for duplication. No part of this package may be included as part of a commercial package without explicit written permission. If you have any questions about commercial distribution of Vivid or its tools I can be contacted at: Stephen Coy Vivid Software 15205 NE 13th Pl. #2904 Bellevue, WA 98007 (206) 641-8615 INTERNET: coy@ssc-vax.boeing.com or uw-beaver!ssc-vax!coy CompuServe: 70413,136 Originally I thought of distributing Vivid via shareware but I seriously dislike the guilt games most shareware authors include. So, I've decided to try beggarware. Here's the product. Use it. If you like it I'd be very happy if you'd sent a check for $30. If you think $30 is out of line, send less (or more) or beer. If you don't think Vivid is worth diddly at least send a postcard explaining why you don't like it. If you want the 386 versions you'll have to register at the $50 level. Sorry about that but times are tough and compilers aren't cheap. Anyway that's about the price of an average game lately so I don't think it's too far out of line. As is usual I will also accept severe sob stories, beer, software (nothing pirated, thank you), barter goods, and young women instead of money. The 386 diskette contains versions of Vivid compiled with both the Zortech and the Intel compilers giving you a choice of DOS extenders to work with. The Intel version is DPMI compliant and has the advantage of supporting virtual memory allowing you to allocate up to 128Mb of disk space as memory for those seriously huge models. The Zortech version is DPMI, VCPI and XMS compatible and takes up less space. It also seems to be able to find about 500K more memory than the Intel version but doesn't directly offer virtual memory support. If another DPMI host which supports virtual memory is active, Windows 3.0 running in 386 enhanced mode for instance, then the Zortech version can take advantage of it. Whether you send anything or not I would like to hear about any suggestions or bugs. Of course requests accompanied by money 9 will tend to get higher priority. I will answer all mail and email even though I may be a bit slow at times. Email tends to get the fastest response. If you have tried to contact me and not gotten a response please try again. If you still don't get a response you might have to resort to US Mail. For those of you into BBSing I can be contacted via any of the ADEnet nodes. ADEnet is a network of BBSs primarily dedicated to providing professional quality support for AutoCAD and other CAD programs. In addition there is also an animation and rendering conference where questions about Vivid, or graphics in general, will be answered. I usually log in on a daily basis so turn around is quite quick. Another BBS I log into a lot is TurboTech. Ray Johnson, TurboTech's sysop and the author of TurboCit, has provided a room for ray trace discussions and has plenty of disk space for images. Check out the conversations, too. The BBS attracts some interesting people. I also frequent CompuServe's COMART forum for those of you with deeper pockets. Ray Johnson's Turbo Tech BBS can be reached at (206)362-6828. Alacrity Design & Engineering Network (ADEnet) ADEnet serves the design and engineering communities by making information useful to them widely available. This is done in three ways: the ADEnet message conferences, the sharing of programs and data files across ADEnet and the dissemination of industry news and press releases. ADEnet Member Listing The following is a listing of the current ADEnet Member BBSs. Each ADEnet Member is an independant BBS operated by a local sysop. Alacrity Software BBS 206-643-5477 Bellevue, WA DUAL (14.4) The Graphics Alternative 510-524-2780 El Cerruto, CA DUAL (14.4) Off Broadway 510-547-5264 San Francisco, CA V32 PC-AUG BBS 602-952-0638 Phoenix, AZ DUAL (9600) Mind Image BBS 612-781-1720 Minneapolis, MN v22, MNP5 LABB 717-394-1357 Lancaster, PA v22, MNP5 [%] AEC-ONLINE 818-360-7022 Los Angeles, CA v22 The University BBS 908-544-8193 Shrewsbury Twp, NJ DUAL (14.4) AZCAD BBS +61-3-481-6873 Australia v32bis Each of the listings has the BBS name and it's (primary) phone number followed by the location. At the end of listing is the highest allowable bits per second (bps) and protocol. DUAL - US Robotics Dual Standard supporting HST and v.32/v.42bis HST - US Robotics Courier HST 10 v32 - Industry standard v.32/v.42bis supporting upto 9600 bps v32b - Industry standard v.32bis/v.42bis supporting upto 14.4k bps v22 - Industry standard v.22 supporting upto 2400 bps MNP5 - Indicates support Microcom Network Protocol Class 5 v42b - Indicates support for CCITT v.42bis [*] CAD/Engineering Services BBS accepts calls between 22:00-05:00 CST. [%] LABB is a multiline system, of which only the main line is listed. The following is a listing of the current ADEnet message conferences. Conference Description -------------------------------------------------------- Autodesk Autodesk product support A/E/C Architectual, Engineering and Construction CAD/CAM Manufacturing, NC Programming Civil Civil engineering CompuGFX Computer graphics, rendering and animation DTP Desktop publishing ForSale Buy & Sell used equipment, job listings Generic Generic Software product support GIS Geographical information services uStation Integraph Microstation support News Industry news and press releases Public Network wide (ADEnet) general conference TechTalk Technical hardware and software discussions Joining ADEnet If you are the system operator of a BBS and interested in joining ADEnet, please download ADENET.ZIP. Within is a more detailed introduction to ADEnet, an application form, and all of the necessary information and data files you'll need. ADEnet currently supports PC-Relay, QWK/REP, and FidoNet's echomail message networking protocols. Alacrity Software BBS serves as the ADEnet hub for nodes using either PC-Relay or QWK/REP. The University BBS acts as ADEnet's "bridge" to nodes using echomail. 11 Acknowledgements First on the list I must thank Mark VandeWettering for his raytracer MTV. Some pieces of Vivid have been lifted directly from MTV's code. Hopefully, I only took the bug-free ones. I'd like to thank Mike Thompson for his various contributions both technical and fermented. Thanks to Gregor Harrison for a lot of help understanding lex & yacc. I still am somewhat confused but I like it that way. Thanks to Jason Osgood of ADEnet for providing a forum for my rantings about Vivid. Thanks to Drew Wells for encouraging me to join CompuServe. And thanks to all those folks who unwittingly helped Vivid through their postings to usenet. Finally, I'd like to thank all of you who took time to send comments (and images and beer and money) for version 1. Coming home to a nice letter and a check in the mail rather than just bills makes the whole week go better. Thank you. GIF is a trademark of CompuServe MS-DOS is a trademark of Microsoft 12 Input File Format One thing to keep in mind while creating Vivid input files is that Vivid's parser is case sensitive. Macros (see preprocessor section) are also case sensitive. Vivid uses a right-handed coordinate system for defining the location of objects in space. A right-handed coordinate system can be visualized as having the x axis pointing to the right, the y axis pointing into the screen and the z axis pointing up. Each object in a scene file will be defined as having a location in space defined by an x, y, z triple in this coordinate system. Colors are defined by an r, g, b triple where each component generally falls in the range 0..1. To make picking colors easier Vivid includes the file color.vc which contains a few dozen predefined colors. Their names and rgb values can be had by viewing or printing color.vc. If this file is #included at the top of you input file these names can be used anywhere an rgb triple is called for. Simple math operations are now supported by the parser. These operations are supported on numbers and on vectors (rgb triples and xyz coordinates). The following is a list of the operations supported. Vectors operations supported include cross product, addition, subtraction, scaling, and dot product. Numerical operations include multiplication, division, addition, subtraction and exponentiation. Sine, cosine, tangent, arcsine, arccosine, arctangent and square root functions are also supported. Vector ops ---------- a b c cross x y z cross product, yields a vector a b c dot x y z dot product, yields a number a b c + x y z vector addition, yields a vector a b c - x y z vector subtraction, yields a vector a b c * n scale a vector by n a b c / n scale a vector by 1/n -(a b c) negate a vector Numeric ops ----------- sin(x), cos(x), tan(x) trig functions asin(x), acos(x), atan(x) sqrt(x) returns square root of x pow(x y) returns x to the yth power *, /, +, - normal mathematical operations It should be noted that these operations introduce some ambiguity to the input language. This problem is aggravated by 13 the fact that the parser can only look one token ahead when it tries to decide how to treat its current input. I encourage you to avoid any such problems by generously using parantheses to eliminate any possible ambiguity in your equations. For example: n * a b c can result in either (n*a) b c or (n*a) (n*b) (n*c) depending on how the parser is feeling that day. If you really want to I'm sure that you could figure out what the parser is doing but that may change in the future so save yourself the trouble and just add the parentheses. Comments can also be included in the input file. Like comments in computer programs, comments in Vivid input files are ignored by the input parser. Comments in Vivid use the same syntax as comments in C++ do. Multi-line comments start with /* and end with */. Anything between the comment delimeters is ignored. Single line comments start with // and end at the end of the line. For example: /* This is a comment which spans multiple lines */ // This is a single line comment Comments should be used as notes in your input files to help remind you or anyone else reading the file what the input is trying to do. Comments can also be used to block out part of the input file while you are setting it up to help speed up test renderings. Normally an input file will be made up of a studio definition which describes the image size, antialiasing and viewpoint followed by the definitions for lights, surfaces, and objects. Object definitions only deal with the geometry of the object. The surface characteristics (color, shine) of the object are taken from the last surface definition preceding the object. 14 Preprocessor To make writing input files easier, Vivid's parser also has a preprocessor. Currently, the preprocessor only supports two functions, the definition of macros and support for include files. Because of the way I implemented the preprocessor Vivid creates a temporary file on the disk called xyzzy.v. (Didn't I already mention this? deja vu) At this time, Vivid does not delete the file after it is done using it. The reason I leave this is that it is sometimes useful to look at when trying to figure out what the preprocessor is doing with your input definition. Feel free to delete it at any time. Macros allow you to associate a name with a string of characters. When the parser sees the name in the input file it will substitute the appropriate string before continuing. A simple example of this is the color definitions in the file color.vc. In color.vc the colors white and blue are defined like this: #define blue (0 0 1) #define white (1 1 1) Once defined you can then use the word "white" wherever you would normally have to type (1 1 1). For example, to create a blue surface with a white highlight you could then do this: surface { diffuse blue shine 20 white } Macro names must start with a letter and may contain letters, numbers, and the underscore character "_". Macro names are case sensitive. Note than in the examples I define the colors with parentheses around the rgb values. This is not required but helps eliminate any potential parser problems as mentioned in the section on the parser's math ability. You can undefine a macro using the #undef keyword. #undef blue If you define the same macro more than once the old values are put onto a stack and the newest one will always be used. If you then undefine the macro, only the newest one will be deleted and the next newest definition will then be active. If you want to define a macro that is longer than one line you have to use the backslash, "\", as a continuation character. 15 #define BLUE_PLASTIC \ surface { \ diffuse blue \ shine 20 white \ } Notice that the last line does not have a backslash after it. Once defined you can then just use the name BLUE_PLASTIC in the input file wherever you would normally type in the whole surface declaration. The other preprocessor directive is #include. This allows you to include other files into your input file. Look at the sample input files and notice that almost the first thing in the file is a line like: #include color.vc This causes the parser to read in the file color.vc which defines a standard set of colors for use in creating input files. Using Dan Farmer's color editor it is quite easy to create new colors to add to this file. The include command is also useful for including objects into your scene. Combined with the transformation commands this will allow you to create objects as seperate files and inlcude them into the scene at any location and orientation. You may also inlcude multiple copies of an object. 16 The Studio The studio structure in the scene file defines all those things that are neither lights, objects or clips. This includes the resolution of the final image, the location of the camera (viewpoint), the direction the camera is pointing, the background color, and various rendering options. Some of the options have default values. These are the values these options will take on if they are left out of the studio definition. The studio definition looks like this: studio { from x y z at x y z up x y z angle ang resolution i j start line stop line aspect asp projection mode ambient acolor background bcolor haze density antialias mode threshold dist jitter aperture size focal_length dist samples n no_shadows no_exp_trans caustics depth max_depth bunching n } "from" is the location of the camera. "at" is where in the scene the camera is pointed. "up" is a vector pointing up, usually 0 0 1. This can be played with to roll the image around the axis of the camera. "angle" is the field-of-view angle given in degrees. 17 "resolution" is the size of the image in pixels, i pixels across and j pixels down. This can also be controlled from the command line using the -i switch. "start" and "stop" allow you to start and stop the rendering at the specified line numbers. I use this for testing new input files and isolating bugs. Those of you with networks might want to consider using this to split up an image for rendering on multiple computers. After the sections are done you can then use paste.exe to glue them together. Currently there is a problem with using start and the resume (-r) flag. Avoid the combination. "aspect" is the aspect ratio of the screen. This is the ratio of width to height of the screen you are rendering your images for. I've found that my Nec 3D has an aspect ratio of about 4/3 and that my Amiga 1084 has an aspect ratio of about 1.2. To determine the proper aspect ratio for your screen measure the width and height of a screen image. The aspect ratio of your screen can be found by dividing the width by the height. Determining the correct aspect ratio of your screen will insure that circles come out looking like circles instead of ovals. Remember, aspect ratio should the the width/height ratio of the displayed image regardless of the image resolution. Together with the resolution these are used to determine the aspect ratio of the pixels. "projection" controls how the 3d world is mapped onto the 2d screen. The default is "flat". This is the standard perspective projection you've all come to know and love. "spherical" projection produces an effect somewhat like a fisheye lens. Things can look pretty strange. Unique to the "spherical" mode is the field of view angle can be greater than 180 degrees. Try a 360 degree panorama some time. The "orthographic" projection mode produces an image where all the eye rays are parallel to each other. For this mode the "angle" parameter has no meaning and is replaced with a "width" parameter. Width defines how wide the screen is in the world coordinate system. Generally, I think that this mode is useless but it was easy to code so why not? The final mode is the "parallax" projection mode. This produces a 2-point projection instead of a 3-point projection like the "flat" mode. In the 18 "parallax" mode, all vertical lines stay vertical on the screen. This was implemented specifically for architectural renderings but is sometimes useful in other contexts. Note that "vertical" is defined by the direction of the up vector. "ambient" is the color of the light that is everywhere in the image. In "the real world" when light strikes diffuse surfaces such as walls, some of it is scattered back into the room. This is why you can still see under a desk even though no light is shining directly underneath it. Most ray tracers, including Vivid, can't handle this diffuse interreflection. But, all hope is not lost. To fake diffuse interreflection Vivid allows you to set an ambient light value. This acts like a light that is shining in every direction at once and does not cast any shadows. For an inside scene values of about .2 .2 .2 seem to work well. Outside scenes look a bit more realistic with a higher ambient value because of the scattering of light the atmosphere does. Most ray traced images that you will see just have the ambient value set to 0 0 0 or black. This produces the sharpest contrasts and gives the image a super-real effect. The default value for ambient is 0 0 0. "background" is the color that will be returned if no objects are hit while tracing a ray. Popular choices are black and sky_blue. If haze is defined then this is the color of the haze. (see below) The background color defaults to black. Alternatively you can do the following: background { palette.map up x y z } This produces a graduated background using the colors in the file palette.map. The file can have any name but is required to have the .map extension. The file is the same format as the img2gif palette files. The first color in the file is the color that will appear in the direction of the up vector. If no up vector is specified then the up vector from the studio structure will be used. "haze" is the density of the fog or haze in the scene. The haze density defaults to 0. This means that there is no haze. A value of .5 means that for 19 every unit a ray travels, half of its color is determined by the background color. This function is exponential, ie if the haze density is given as .5 the color of a ray going 1 unit will be .5 times the color of the object it hits and .5 times the background color. A ray going 2 units will be .25 the color of the object and .75 times the background color. For most images this parameter can be ignored and the default value of 0 used. "antialias" determines whether or not antialiasing is performed and what type is used. This can also be controlled from the command line using the -a switch. The valid modes are: none -- Do one ray per pixel, right through the center. Results are blocky but relatively quick. This is the default. quick -- This does a subsampling approximation of the image. In areas of even color the most speedup is gained. This is the fastest mode but the results are not useful for much more than test images. At best (ie on a blank image) this should be about 15 times faster than the above mode. In general I notice about a 3-5 times speedup. This is a great mode for doing test renderings. corners -- Shoot a ray at each corner of the pixel and average the results. Since the corners are shared by adjoining pixels this means about one ray per pixel. The results are almost as quick as none but usually have a better look. Effectively this is the same as running a smoothing filter over the image. adaptive -- Rays are shot at the corners of the pixel. If they are within a certain threshold of each other the program moves on to the next pixel. If they differ by more than the threshold value, the pixel is subdivided into four subpixels and sampled again. The corners of the subpixels are then compared against the threshold and if they are still too far apart the are subdivided once more. The effective result of this is that in areas of constant or smoothly changing intensity only one ray per pixel is shot. At edges or other sharp color transitions up to 25 rays per pixel may be averaged to determine the color of the pixel. The result is fairly good antialiasing without too much undo overhead. "threshold" is the threshold value used by the adaptive 20 mode of antialiasing. The default threshold is 16. Valid values are 0..255. This parameter also affects the quick mode. In general, lower values will produce better results but take more time. "jitter" is a flag telling the system to add a little bit of randomness to the direction each ray is shot. Combined with antialiasing this helps to break up the patterns sometimes caused by sampling an image on a regular grid. Images with regular patterns such as checkerboards disappearing into the horizon will benefit most from jitter. "aperture" is an optional parameter which allows the ray tracer to model a more realistic camera. The default aperture is 0 which models a pinhole camera. With an aperture greater than 0 objects at the focal length (see below) will appear in sharp focus while objects nearer or further from the viewpoint will be blurred. The larger the aperture, the more exaggerated the blurring will be. Using this option will greatly increase the amount of time needed to generate an image because Vivid uses distributed ray tracing to model the effects of a camera with a non-zero aperture. This causes the number of rays necessary to calculate the color of a pixel to increase greatly. The default is to shoot 8 rays instead of one ray whenever aperture is greater than zero. This value can be controlled with the "samples" parameter below. "focal_length" determines the distance from the camera to the focal plane where objects are rendered in focus. This option is used in conjunction with the aperture option. Objects which are a distance equal to the focal length away from the camera will be in sharp focus. The default for the focal length is the distance between the "from" and "at" points which determine the viewpoint and the viewing direction. "samples" controls the number of rays shot when a non-zero aperture is used. The default is 8. "no_shadows" causes all shadow calculations to be turned off. The speed increase gained by turning shadows off is especially useful when doing test images of a new scene. The can also be controlled using the command line switch -n. "no_exp_trans" is a wonderfully intuitive name for a flag 21 that turns off the exponential attenuation of the rays as they pass through transparent objects. Got that? Let me try again. Normally when Vivid shoots a ray through a transparent object (glass) the color of the ray is tinted by the color of the glass and is a function of the distance that the ray has to travel through the glass. For example if you have two sheets of coke-bottle green glass where one is 1/4 inch thick and the other is 2 inches thick, light passing through the thicker one will be darker. The relationship between the thickness and the amount of tinting is exponential. This causes problems with single sided glass because when Vivid tries to compute the thickness of the glass the distance calculated is from the glass to whatever wall or floor the ray hits next. Hence the windows will tend to be way dark. When you set the no_exp_trans flag in the studio structure Vivid only uses the transparent color of the surface to calculate the tint and totally ignores the distance that the ray travels. This tinting also affects shadow rays. "caustics" is an experimental flag which turns on Vivid's faked caustics. Caustics are those neat patterns of light that are produced as light passes through a transparent object. I've been playing around with a couple of ways to get some of the effect without having to spend a few days per image doing the real thing. The effect is pretty subtle but does seem to make some images look better. Joe Bob says check it out. "depth" lets you limit the maximum recursion level to which rays will be traced. At a depth of 1 only eye rays are traced. A depth of 2 will trace 1st level reflections and refractions. The maximum value allowed is 20. This is also the default value. This can also be changed using the command line switch -d. "bunching" allows you to control the branching factor of the tree created by the bounding boxes. I'm not really sure what an optimal value is but the default value of 4 seems to work well for most cases. At any rate, values less than two are guaranteed to cause grief. Higher values will cause the bounding box tree to branch more at each level and therefore be shallower. Lower values will do the opposite. I find that experimenting with takes more time than I ever save so the only time I use it is when I am close 22 to running out of memory. Using a higher value like 8 or 12 will create fewer composite (bounding box) nodes in the tree and save soom memory. Feel free not to ignore it without fear of missing out on something. 23 Lights Lights come in four flavors, point, directional, spherical and spot. Just as the name implies a point light is a light source occupying a single point in space. It has position, color, and attributes determining how much the intensity of the light falls off with distance. A directional light acts like a point light source infinitely far away with no reduction in intensity over distance. A spherical light source actually has a radius to it and can provide shadows with penumbra (soft edges). This feature, however, adds a considerable amount to the time needed to render the image. Spot lights produce a cone of light that falls off on the edges. These are quite nice for highlighting areas of your model. In most cases they also produce the least number of shadow rays making them quicker than just putting a point light inside a cone. The definition for a point light source looks like this: light { type point falloff f // defaults to 0 position x y z color r g b } The falloff parameter determines how the intensity of the light is reduced over distance(dist). In the real world the light intensity falls off as 1/(dist*dist) (f=2). Vivid also allows the light to fall off as 1/dist (f=1) and not to fall off at all (f=0). Why would you want to use anything except f=2? Simplicity is one reason. With f=0 you can set the light's color to 1 1 1 and know that whatever objects the light falls on will be illuminated fully regardless of the distance of the light from the object. With f=2 you must take into account this distance. If the object you wish to be illuminated is 3 units away from the light then in order to get the same amount of illumination that f=0 provides you must set the color to 9 9 9 ie 3^2. For f=1 the color would have to be 3 3 3. In the real world much of the light around us does not come directly from the light source. It often bounces off of other objects on its way to the object we are interested in. Since Vivid, like most ray tracers, does not model this interobject diffuse reflection we can achieve much the same effect by have the light intensity fall off linearly with distance, ie f=1. The default value is f=0. The definition for a directional light source looks like: light { type directional 24 color r g b direction dx dy dz } or the direction can be replaced by a from and at pair: light { type directional color r g b from x y z at x y z } The direction vector points along the direction the light is travelling. Since the light is assumed to be at infinity, there is no falloff parameter. If you are having difficulty understanding how the direction parameter works is is sometimes useful to note that direction x y z is the same as center 0 0 0 at x y z. The definition for a spherical light source looks like: light { type spherical position x y z radius r color r g b falloff f samples n } Spherical lights differ from point lights in that the shadows they cast have penumbra. Normally when a ray hits a surface a shadow ray is shot toward each light. If the shadow ray hits any surface on the way to the light then that light is blocked and the surface is in the shadow of the blocking object. With spherical light sources multiple shadow rays are shot. Each one is shot to a random point within the radius of the light. If the light is half blocked by an object, approximately half the shadow rays will be blocked and half will pass through to the light. The ratio of blocked to not-blocked shadow rays is then used to determine how strong the shadow is. As you might expect, the extra shadow rays will add a lot of extra time to the rendering. Some references refer to these light sources as extended light sources. The number of shadow rays shot each time is controlled by the samples parameter. The default value for this is 16. The definition for a spot light source looks like: 25 light { type spot position x y z direction dx dy dz min_angle angle1 max_angle angle2 color r g b falloff f } Like the directional light, the direction parameter may be replaced with the at x y z to specify where the light is shining. Min_angle and max_angle define the shape of the cone of light produced by the spot light. Everything within the min_angle angle of the axis of the light will be fully illuminated. From there the light intensity will fall off until max_angle is reached. For example if you want a cone of light 30 degrees wide with sharp edges you would define min_angle and max_angle to be 30. To get the same size light but one that fades out at the edges you would define max_angle to be 30 and min_angle to be 0. Each light source can also have a couple of other parameters. They are no_shadows and no_spec. As you've probably guessed, these allow you to turn off shadows and specular highlights for each light. Used together with a directional light source of low intensity (.2 .2 .2) this is often a nice alternative to global ambient light. Global ambient light tends to make objects appear flat whereas this technique will provide subtle shading without the shadows and spots of your "real" light sources. One thing to note is that even if a light is within the viewing scene it will not appear as an object. If you want you lights to be visible you can wrap a transparent shell around them using a sphere. Example: // Define a point light at 2 3 4 that shows up in the scene // as a light with radius 1. light { center 2 3 4 type point color white } // glass shell surface { ambient white // same color as the light transparent white // totally transparent } sphere { 26 center 2 3 4 radius 1 } For details on spheres and surfaces skip ahead. 27 Surfaces Surface structures allow you to define the surface characteristics of the objects you are rendering such as color, reflectivity and texture. When a surface is defined in the input file, it is applied to all the primitives following it until a new surface is defined. This allows multiple objects to be entered without having to repeat the surface characteristics. A simple surface structure looks like this: surface { diffuse r g b // defaults to 0 0 0 (black) ambient r g b // defaults to 0 0 0 specular r g b // defaults to 0 0 0 shine pow // defaults to 0 transparent r g b // defaults to 0 0 0 ior num // defaults to 1.0 fuzz magnitude // defaults to 0.0 no_antialias // turn off antialiasing } All of these components of the surface are figured in when the ray tracer determines the color of an object in the scene. The diffuse color is the actual color of the object as seen when illuminated by a full white light. A value of 0 0 0 signifies a black object while a value of 1 1 1 indicates white. The brightness of this component depends on the amount of light falling on the surface at that point. The ambient term is sometimes also referred to as the self-luminous component. This is the color the object will appear in a totally dark scene. The specular component specifies the reflectivity of the surface. A value of 1 1 1 will produce a mirror-like reflection. The shine value determines how large the specular spot will appear on a surface. Low values, 1..10, will produce large, soft-edged specular highlights while high values, 1000 or more, will produce a small, sharp spot. Traditionally the brightness and color of the spot is in direct proportion to the specular component. The problem is that sometimes it would be nice to have a blue object with specular highlights without having the extra overhead of tracing reflected rays. Therefore Vivid allows a second form for defining specular spots: shine pow r g b In this case the color given will be used instead of the specular component of the surface. The transparent component allows you to define how transparent the surface is. A value of 1 1 1 will appear glass-like because it allows all colors to pass through while a value of 1 0 0 will produce a surface like red 28 glass since it only allows red light to pass through. A surface with a transparent component of .9 .9 .9 will appear partially transparent with the amount of light passed through based on the thickness of the object the light is passing through. The index of refraction, ior, determines how much the ray is bent as it passes into the transparent surface. In reality this is related to the relative density of the surface. To simulate glass values of about 1.1 to 1.3 seem to work best. The ior of diamond is 2.6. Fuzz is a way of adding random noise to the surface normal of the object when its color is determined. Since the diffuse color of the object is affected by the angle the light hits the surface this randomization can produce a sort of coarse texuture to an object. Applied to mirrored or transparent surfaces this produces an affect much like frosted glass. Generally, small values of fuzz, .01 to .3, seem to work best. The no_antialias flag tells the adaptive antialiasing to effectively turn off for that surface. In general this is not something that you want to do except in a few special cases. The original reason for this parameter was that fuzzy surfaces can cause the adaptive antialias option to shoot lots of rays and slow down the image generation considerably. By adding the no_antialias flag to the surface definition you still get the benefits of the adaptive antialiasing along the edges of the objects but you avoid the slowdown that can be caused by any large, fuzzy surfaces. Note, however, that this will change the look of the surface. Try cutting the amount of fuzz in half when using this option to preserve the amount of color variation in the surface. As an aid to those of us that can't type, some keywords may be abbreviated: surf, diff, amb, spec and trans. I trust that you'll be able to figure out what goes with what. Some examples are probably in order. // simple red surface surface { diff 1 0 0 } // self-luminous blue surface { ambient 0 0 1 } // mirror with specular highlights surface { spec 1 1 1 shine 100 } // glass with some reflection 29 surface { spec .3 .3 .3 shine 30 trans .7 .7 .7 ior 1.2 } In general, the rule of thumb is that amb+diff+spec+trans should be less than or equal to 1 1 1. Of course since we are defining our own universe anything is possible. 30 Bump Mapping Bump mapping is a means of giving a surface some texture which enhances the realism of the surface by adding ripples or bumps to the surface. The surface isn't actually distorted but its normal is. This allows a simple surface to appear very complicated. Bump definitions are included inside of the simple surface definitions, ie: surface { diffuse red bump { ... } } A sample wave bump map looks like this: bump { wave { center 1 2 3 wavelength 2.0 amplitude 0.2 damping 0.9 // defaults to 1.0 phase 0.0 // defaults to 0.0 } } Center defines the source of the wave. Wavelength defines the crest to crest distance of the wave. Amplitude defines the maximum amount that the surface normal is bumped. Values under 1 are definitely best. The damping parameter defines how much the amplitude falls with distance. In the example given the amplitude will decrease by 10% for each wavelength of distance from the source. The phase is a number between 0 and 1 which defines a starting offset for the phase of the wave. This can be used in animations to create a wave which appears to move realistically by incrementing the phase by a small amount for each frame. More than one wave may be defined within the bump structure. By defining three or four wave sources at various locations with differing wavelengths and amplitudes a very realistic rippled surface can be created. 31 Turbulence can also be used to perturb the normal of a surface. The definition of turbulence looks like this: bump { turbulence { scale 1 1 1 offset 0 0 0 amplitude .5 terms 4 } } The turbulence function takes the location of the ray intersection and returns a random number in the range +- amplitude. The scale and offset factors are applied to the xyz location before the turbulence function is called. The terms parameter allows you to build a fractal-like surface. When terms>1 the turbulence function is summed multiple times. Each successive term in the sum has its scaling doubled and the amplitude halved. This produces the varying levels of self-similarity associated with fractals. The sample file vivid1.v uses this feature and a non-symmetric scaling to produce the "spun-chrome" look on the large sphere. Turbulence and wave definitions may be included with each other inside a bump definition. 32 Textured Surfaces Vivid also allows the use of solid texturing to enhance the realism of the surfaces rendered. Textured surfaces are really just two surfaces with some definition of which appears where on the object and how the two surfaces are blended together. The surfaces can be "layered" in one of three patterns checker, spherical or noise. The checker pattern produces a 3-d checker. The size of the blocks is controlled by the scale factor. If one of the scale parameters is 0 then the pattern is assumed not to change along that axis. An example would be a simple checkerboard pattern with checkers 2 units on a side colored black and white. This pattern is defined to extend infinitely along the z axis. surface { texture { pattern checker scale 2 2 0 offset 0 0 0 // default fuzz 0 // default blend 0 // default surface { diff black } surface { diff white } } } A scale of 2 0 0 would create a pattern which changes every 2 units in the x direction but is continuous in the y and z directions. This is equivalent to 2-unit thick slabs of material stacked along the x axis. The spherical pattern produces concentric layers of alternating surfaces. When one of the scale parameters is 0 concentric cylinders are formed with the axis of the cylinders along the zero'ed axis. This is useful for wood textures. The spherical pattern also requires 2 radius definitions for the layers. The first radius is used for the first surface, etc. surface { texture { pattern spherical scale 2 2 0 radius 1 radius 2 surface { diff black } surface { diff white } } } 33 The noise pattern uses the output of the noise function directly to pick between the two surfaces. This is useful for producing textures like granite. By using unequal scaling values in the x y and z directions you can get a streaked looking surface. I've also used this using tan and brown surfaces to produce a fine wood-grain look. surface { texture { pattern noise terms 4 scale x y z surface { diff white } surface { diff black } } } The fuzz and blend parameters may be used to soften the edges between the two surfaces. Their values range from 0 to 1. The blend parameter produces a smooth transition between the surfaces. The value of the blend parameter determines the width of this transition area. The fuzz parameter adds noise to the point being checked in proportion to its value. This produces a coarse, speckled transition between the surfaces. The turbulence function mention in the bump map section may also be applied to textured surfaces. By varying the parameters the effect can be made to range from a slight perturbation of the pattern, to a marble look, to excessive turbulence. A simple example is presented below. Due to the way Vivid's parser works if you want to bump map a surface which is also textured the bump definition must appear in the surface structure before the texture definition. Also notice that the surfaces defined in a texture definition need not be simple surfaces. They may also be textured. The following is a definition for a checkerboard surface with checks 10 units on a side. Half the checks are black while the other half have a red/white marble finish. surface { texture { pattern checker scale 10 10 0 surface { diff black } surface { texture { pattern checker scale 1 0 0 blend 0.7 turbulence { 34 amplitude 3 terms 4 } surface { diff white } surface { diff red } } } } } 35 Vivid also allows a Mandelbrot pattern as demonstrated in the sample input file vivid1.v. The Mandelbrot pattern was put into Vivid as a joke one evening. Currently the format for specifying it is terrible. The first two parameters of the scale value determine the x y scaling of the set while the 3rd parameter determines the maximum number of iterations. The result of the iteration calculation is then scaled by the max number of iterations to determine the relative percentages of the surface parameters for that point. OK, the way it works is like this. The actual point of intersection, x y z, is tranlated using the scale and offset values before being put into the Mandelbrot calculation. new_x = x * scale_x + offset Like I said, this is about a backwards as I could make it and still have it work. Since the point is multiplied by the scale rather than divided the pattern gets bigger as the scale value gets smaller. Normally the Mandelbrot set extends from -2 to +1 along the x axis (more or less). If you want to get it to fit from -20 to +10 you would have to scale it by 0.1 in the x and y axes. Stupid? Yes. Now, since the offset is added on after the multiplication that makes it work in the Mandelbrot coordinate system rather than the world coordinate system. Continuing the example above, if you wanted to move the center of the set to world coordinates 300, 100 you would have to give an offset of 30 10. Using macros I think you can alleviate most of this garbage. #define ITERATION_LIMIT (64) #define SCALE (10) #define X_OFFSET (300) #define Y_OFFSET (100) surf { texture { pattern mandelbrot scale (1/SCALE) (1/SCALE) ITERATION_LIMIT offset (X_OFFSET/SCALE) (Y_OFFSET/SCALE) 0 // etc... } Now all you have to do is change the defines and things will work much more like you expect. Once I figure out a reasonable way of doing this Mandelbrot and Julia sets will become a real part of Vivid. (Yeah, I know I said that last time but it's still true.) 36 Sphere The sphere is the simplest of the primitives supported by Vivid and generally the fastest to perform an intersection test with. The format for a sphere is: sphere { center x y z radius r } where x y z is the location in space for the center of the sphere and r is the sphere's radius. As an example here is the definition for two glass spheres, one which is hollow and one which is solid. // glass surface surface { trans 1 1 1 shine 200 1 1 1 ior 1.2 } // solid globe sphere { center 1 0 0 radius .9 } // hollow globe sphere { center -1 0 0 radius .9 } // outer surface sphere { center -1 0 0 radius .8 } // inner surface Just for laughs vivid also includes a fuzzy spheres option. These spheres appear fuzzy because they have no fixed radius. To define a fuzzy sphere define a normal sphere and add a fuzz parameter. This defines how much larger the radius will randomly be. Each time the ray tracer performs an intersection test with the fuzzy sphere, the radius to test against is randomly chosen to lie between the radius and radius+fuzz. // fuzzy sphere with radius between 0.5 and 2 sphere { center 0 1 2 radius .5 fuzz 1.5 } 37 Ring The ring primitive may also be described as a washer or disk. The definition for a ring consists of a location, a surface normal, and a minimum and maximum radius. The minimum radius may be zero producing a disk without a center hole. Because the intersection for the ring is faster than for a polygon the ring is a good choice for use as a ground plane underneath the objects you are ray tracing. The format for the ring definition is: ring { center x y z normal a b c min_radius r0 max_radius r1 } The surface normal a b c does not have to be normalized. If you just want a disk without a center hole the min/max radius definitions may be replaced with a single radius definition as follows. ring { center x y z normal a b c radius r } Whoa, just got a clue that not everyone knows what a surface normal is. A surface normal is a vector that is perpendicular to a surface, ie one that points straight out from the surface. For example, the surface normal for the floor in you room would be a vector pointing straight up into the air. The surface normal of your monitor is the vector pointing straight out between you eyes. Simple enough? Ok, now a normalized vector is one which has been scaled to have its length equal exactly 1. This is usually done by calculating the length of the vector then dividing each of the vector's components by the length. Vectors of length 0 cause no end of problems. 38 Polygon Polygons may have any number of vertices (well, a minimum of three is required). The vertices must all lie within the same plane otherwise the results will be strange. The order of the vertices may be either clockwise or counter clockwise. polygon { points 4 vertex 1 1 0 vertex 1 -1 0 vertex -1 -1 0 vertex -1 1 0 } This will produce a square polygon 2 units on a side centered at the origin with a surface normal equal to 0 0 1. 39 Triangular Patch The triangular patch is useful for building objects with complex shapes that you want to appear smooth. The patch is defined by three vertices and explicit surface normals for each vertex. In general, manually entering in patches will probably be too tedious to be of much use but when using computer generated input files the results will be worth the extra programming effort. patch { vertex 1 0 0 normal .1 0 1 vertex 0 1 1 normal 0 .1 1 vertex 0 0 .5 normal -.1 -.1 1 } 40 Cone The cones implemented by Vivid are actually truncated cones. They have a radius both at their base and at their apex. When these radii are equal the cone does a fair imitation of a cylinder. To get a pointed cone enter 0 for one of the radii. cone { base 1 1 1 base_radius 4 apex 0 0 5 apex_radius 1 } Rings are useful for putting caps on the ends of cones. Even for a cone at an odd angle the position and normal of the ring can be easily calculated. To cap the apex end of the cone the ring's center is equal to the apex, the ring's radius is equal to the apex_radius and the ring's normal is equal to apex-base. Using the example above the definition for a ring to cap the apex end of the cone would look like this: ring { center 0 0 5 radius 1 normal -1 -1 4 } 41 Transformations Vivid's transformation commands allow you to move and scale objects. Transformation commands apply to all the primitives following them until they are "popped" from the transformation stack. The format for a transformation command is: transform { scale s rotate x y z translate dx dy dz } Scale changes the size of the objects. Scale may also be defined as a vector to create objects that are scaled differently in each axis. Unfortunately this doesn't work with all primitives, only polygons and patches can be scaled non-uniformally. The rotate command rotates the object the given number of degrees around each axis. The translate command moves the object. Any of these may be left out or used more than once. They can also be used in any order and will be applied to the objects in the order that they are given. It is very important that you get the order correct. An object that is rotated 90 degrees around the Z axis and translated 10 units along the X axis will end up at 10 0 0 with a 90 degree twist whereas if the operations are applied in the other order the object will end up at 0 10 0. Sometimes it helps to play around with real objects a bit and work through some of the transformations first. Remember that all rotations are done around the axes, not necessarily around the center of the object. This should also be kept in mind when building new objects. Put 0 0 0 at the objects "natural" center of rotation. This will help greatly when building scenes with the objects. For example, the natural "center" for a car model would be at ground level in the center of the car. This allows the car to fairly easily be placed in the scene where you want it. To remove a transformation from the transform stack use the "transform_pop" command. Sometimes you will want to nest transform commands. This is useful for creating multi-part objects that move relative to each other but also need to move as a whole unit. For example, say you want to create a tank model with a turret that you can rotate. Assume that the body of your tank model is in the file tank.vo and the turret is in turret.vo. To place the tank in your scene, your input file would look like this: #define TURRET_ANGLE (30) // rotation for turret transform { translate x y z } // move whole tank 42 #include tank.vo // include body geometry transform { rotate 0 0 TURRET_ANGLE } #include turret.vo // include turret geometry transform_pop // clean up transform stack transform_pop Using this technique complicated models can be built and positioned with relative ease. There is currently one major drawback to using transformations, surface textures don't move with the object. This isn't too big a deal for single frame images but will make animating textured objects look very poor. Yes, I'm working on it. 43 Clipping Primitives can also be "clipped" to produce more complicated shapes. Basically the way a clip works is that a primitive is defined with a clipping surface which cuts off part of that primitive. For example a sphere can be clipped against a plane to produce a hemisphere or a cone can be used to clip a hole through another cone. There are three types of clipping surface: planes, spheres and cones. Clips are defined within a primitive's definiton. You may have more than one clip per primitive. Clips also transform along with their primitives. A clipping plane is defined by a point and a normal. clip { center x y z normal x y z } The part of the primitive on the normal side of the plane will be kept while the part on the other side will disappear into the Ronald Reagan Memorial Library. (You may think I'm lying but you'll never really _know_ until you go check.) For example, if you want to get a hemipshere of radius 1 centered at the origin if would look like: sphere { center 0 0 0 radius 1 clip { center 0 0 0 normal 0 0 1 } } Note that the clip's normal is pointing upward. This will give you the top half of the sphere. If you change the normal to 0 0 -1 you will get the bottom half. Clipping spheres are defined as: clip { center x y z radius r inside or outside } With a clipping sphere you can choose to either keep the part of the primitive inside of the sphere or the part of the primitive outside of the sphere. You may have already guessed this but that's why the inside and outside keywords are there. Clipping cones look like: 44 clip { apex x y z apex_radius r base x y z base_radius r inside or outside } Just like the cone primitive you may also just define a single radius to get a cylinder. Sometimes you will want to apply the same clips to a group of primitives. To do this define global clips using the global_clip keyword: global_clip { clip { ... } clip { ... } clip { ... } } The clip_pop keyword will cause the previous section of clips to be popped off the stack much like the transform_pop does for transformations. 45 Support Programs img2gif.exe Once vivid has completed an image you need some way to view the results. That's where img2gif comes in. Img2gif is used to convert the raw img file that Vivid outputs into a gif file. The resulting gif file can then be viewed using your favorite view program. My favorite is Brad Montgomery's vpic. Look for it on you local BBS. In a pinch, even fractint can be used to view gifs. When running img2gif you have 2 main decisions to make: a) how should the program pick the palette and b) should the program do any dithering of the output. Picking the palette is not always an easy problem. Standard VGA allows us to view only 256 of the available 262,144 colors. An "average" lores image will have about 7000 colors which we need to represent with our palette of 256. A higher resolution image my have tens of thousands of colors to represent. The following algorithms are currently implemented in img2gif. Palette File -- A palette may be input via a .map file. This uses the same format as .map files created/read by fractint. This option is useful for creating multiple gifs with the same palette for animation. Popularity -- The popularity algorithm chooses the palette by picking the 256 colors which represent then greatest number of pixels in the image. The remaining colors in the image are mapped onto the palette using a minimum distance formula. The popularity algorithm tends to work best on images with a lower number of colors. One typical problem of the algorithm is that small highlights may not be colored correctly. Highlights generally only cover a few pixels so their color usually doesn't have enough representation in the image to be chosen by the popularity algorithm. To help alleviate this problem, img2gif forces the corners of the color cube (white, red, green, blue, cyan, magenta, yellow, and black) to be selected as the first eight palette entries. Since most highlights are white this greatly helps reduce unwanted artifacts in the resulting image. Median Cut -- The median cut algorithm tries to choose a more well balanced set of colors to represent the image. The general idea of the median cut is to choose a palette in which each entry represents about the same number of pixels in the image. This helps to correctly color highlights that a pure popularity algorithm might miss. Img2gif also allows you to limit the number of pixels represented by any one color. This, in effect, increases the importance of the colors in highlights and other small regions of the image. With this limit set to one, every color in the image is given the same weight without 46 regard for the number of pixels it covers. Fixed Palette -- The fixed palette option uses a predetermined palette instead of choosing one based on the content of the image. This has the advantage of being much faster. When choosing by popularity or median cut img2gif must first build a tree structure in memory containing every color in the image and a count of the number of pixels represented by that color. After choosing the palette the colors in the tree must then be mapped onto the colors in the palette. This can slow img2gif down quite a bit. When using a fixed palette the image colors are mapped directly onto the palette colors via simple equations eliminating the need for the tree structure and the costly remapping of the images colors to an arbitrary palette. Also, the current version of img2gif can only support a tree containing about 57,000 colors. After this my clone runs out of RAM. The fixed palettes supported by img2gif are: 0 : A grey scale image using 256 grey scale tones. Due to VGA limitations this will display as 64 shades on most PC systems. 1 : Divides the color cube into 8 shades of red, 8 shades of green, and 4 shades of blue. Generally produces fairly bad images but kept in for nostalgia. Blue was chosen to get the lower resolution because in general the human eye is much less sensitive in the blue range and has a harder time focusing on edges therefore the loss of color resolution is not missed as much. 2 : Divides the color cube into 6 shades of red, 7 shades of green, and 6 shades of blue. This option gives the best balance of speed vs good color representation. I've found that it works best with images that have a large number of colors. The addition of dithering is usually helpful. Dithering is the process of displaying pixels of differing colors next to each other in such a way as to produce the illusion of more colors. An every day example of this is your television. Stick your nose up to the picture tube and notice that the screen is made up of tiny red, green, and blue dots which vary in intensity. These are the only colors your TV actually produces but your eye combines them to produce the rest of the spectrum. Well, at least the spectrum according to NTSC. Floyd Steinberg dithering is the original error diffusion dithering algorithm. Error diffusion dithering attempts to compensate for a limited palette by insuring that the sum of the errors in any region of the image is zero even though the colors at each individual pixel may contain some error. At each pixel as img2gif scans through the image the nearest color for that pixel is chosen. Generally this color will not be exact. The difference between the chosen palette color and the true color is 47 then added to the neighboring pixels which have not been scanned yet. Floyd Steinberg dithering has the advantage of having a "random" look to it. Generally, it does not produce noticeable patterns in the output image. Ordered dithering works by adding a fixed amount to each pixel based on its location. The sum of these additions over a region of the image is equal to zero. Since ordered dither does not take into account the values of neighboring pixels it produces fixed patterns in the images which may be distracting in some cases. Ordered dithering combined with a fixed palette has an advantage if you are using Vivid to produce single fames for an animation. During an animation the background (non-moving) regions of the image will stay constant whereas if you use Floyd Steinberg dithering the patterns constantly shift around and below any moving objects on the screen. The random noise option allows you to add white noise to an image. In some cases this may help as much as dithering. It can also be used in conjunction with either of the dithering options. Interesting effects may be obtained by increasing the range of the noise. Images converted with a large amount of noise tend to take on a distinctly grainy look. Interesting, but not overly useful. The default range for the noise is +-8 on a 0..255 range. This value was chosen strictly because it seemed to look good in most cases. Random noise has the advantage of being constant from image to image. This means that smooth areas in a series of images will not shimmer during animation. Ok, fine, now how do I run img2gif? Well, it goes like this: img2gif [-m #] [-f #] [-p palfile] [-d] [-o] [-r [#]] file The default palette choosing algorithm is the popularity method. [-m #] chooses median cut and limits each color to represent at most # pixels. # defaults to 64K. [-f #] chooses fixed palette #. # defaults to 0, grey scale. [-p palfile] reads the palette out of the file palfile.map. If this file does not exist the palette will be chosen by median cut, popularity, or fixed depending on what other flags are set. After the palette is chosen it will then be written to the file palfile.map. [-d] chooses Floyd Steinberg dithering. [-o] chooses ordered dithering. [-r #] chooses random noise +- #. # defaults to 8. 48 For example: img2gif -f 2 -d test will convert test.img to test.gif using fixed palette 2 and Floyd Steinberg dithering. Assuming that test.map does not exist img2gif -m -p test test1 img2gif -m -p test test2 img2gif -m -p test test3 img2gif -m -p test test4 will convert test1 using median cut to choose the palette. The palette will then be saved to test.map. Test2, test3, and test4 will then be converted using the palette in test.map. In general the best results will be obtained by using the -m -d flags as in img2gif -m -d test. For images that tend to be fairly monochromatic I've found that -m -d -r 4 works well. Adding the small amount of random noise helps break up some of the banding. On the down side it also adds to the .gif file size since it is tough to compress random noise well. To get nice looking grey scale images try -f 0 -r. 49 paste.exe The paste tool will allow you to join together two images in raw file format either vertically (one over the other) or horizontally (side by side), ie. two 320x200 images may be joined together to form a 320x400 image or a 640x200 image. In order for two images to be joined vertically the images must have the same width. In order for them to be joined horizontally they must have the same height. The calling sequence for paste is: paste [-h | -v] in_file1.img in_file2.img out_file.img In the horizontal case in_file1.img will appear to the right of in_file2.img. In the vertical case in_file1.img will appear above in_file2.img. 50 up.exe Up allows you to double an image's size vertically, horizontally, or in both directions. The increase in size is done by pixel duplication, no interpolation is used. The calling sequence is: up [-h | -v] in_file.img out_file.img If no flag is given, up increases the image's size in both directions. For example if test.img is a 320x200 image up -v test.img test2.img will create test2.img with a resolution of 320x400. 51 down.exe Down does the opposite of up. (surprise!) When reducing an image's size, down averages the appropriate pixels to produce the new pixel. The calling sequence is: down [-h | -v] infile.img outfile.img If no flag is given, down reduces the image in both directions. For example if test is a 320x200 image down test.img test2.img will produce test2.img with a resolution of 160x100. 52 File Formats The information included here about the file formats is to aid people in writing their own image manipulation utilities compatible with the Vivid formats. I encourage you to experiment and share any interesting results with the rest of us. img -- This is the format of the files output by Vivid. The images are stored 24-bits per pixel with a simple run length encoding scheme to help keep the size down. The run length encoding works by replacing a repetitive string of the same color with a count value and that color written only once. "Runs" of the same color are not allowed to continue beyond the end of the scanline to make working a scanline at a time easier. The format consists of a 10 byte header followed by the image data. The 16-bit numbers in the header are stored most significant byte first. This format is compatible with that used by Alias Research. (I think.) <2-bytes> x size of image <2-bytes> y size of image <2-bytes> first scanline, usually 0 <2-bytes> last scanline, usually y size - 1 <2-bytes> number of bitplanes, always 24 The image data format looks like: <1-byte> a repeat count for the following color <1-byte> blue, 0..225 <1-byte> green, 0..255 <1-byte> red, 0..255 This is repeated as many times as necessary to complete the image. Note: runs do not wrap from one scan line to the next. This helps simplify post-processing. It has been noted (hi Ray!) that in some of Vivid's antialiasing modes an extra scan line is output to the file. This may eventually get fixed but for now the best thing to do is to always use the info in the header rather than looking for the end of file. map -- This is the format used for palette files. This format was chosen to be compatible with fractint. The file is in plain ASCII text and consists of 256 lines each containing the red, green and blue value for that palette entry. The rgb values are integers in the range 0..255. 53 Bibliography For anyone interested in ray tracing and how it works the following books are highly recommended. Further references may be found in the bibliography at the end of Glassner's book. "Computer Graphics, Principles and Practice, 2nd Ed.", Foley, van Dam, Feiner and Hughes, 1990, ISBN 0-201-12110-7 "An Introduction to Ray Tracing", Andrew S. Glassner, ed., Academic Press, 1989, ISBN 0-12-286160-4 "Illumination and Color in Computer Generated Imagery", Roy Hall, Springer-Verlag, 1989, ISBN 0-387-96774-5 54 Index #define. . . . . . . . . . . . . . . . . . . . . 15 #include . . . . . . . . . . . . . . . . . . . 5, 16 ambient. . . . . . . . . . . . . . . . . . . . . 19 angle. . . . . . . . . . . . . . . . . . . . . . 17 antialiasing . . . . . . . . . . . . . . . . . . 20 aperture . . . . . . . . . . . . . . . . . . . . 21 aspect ratio . . . . . . . . . . . . . . . . . . 18 at . . . . . . . . . . . . . . . . . . . . . . . 17 background color . . . . . . . . . . . . . . . . 19 bibliography . . . . . . . . . . . . . . . . . . 54 blend. . . . . . . . . . . . . . . . . . . . . . 34 bump mapping . . . . . . . . . . . . . . . . . . 31 bunching . . . . . . . . . . . . . . . . . . . . 22 caustics . . . . . . . . . . . . . . . . . . . . 22 changes for 2.0. . . . . . . . . . . . . . . . . 3 checker pattern. . . . . . . . . . . . . . . . . 33 clipping . . . . . . . . . . . . . . . . . . . . 44 clipping plane . . . . . . . . . . . . . . . . . 44 colors . . . . . . . . . . . . . . . . . . . . . 13 command line flags . . . . . . . . . . . . . . . 6 comments . . . . . . . . . . . . . . . . . . . . 14 compound surfaces. . . . . . . . . . . . . . . . 33 cone . . . . . . . . . . . . . . . . . . . . . . 41 coordinate system. . . . . . . . . . . . . . . . 13 depth. . . . . . . . . . . . . . . . . . . . . . 22 dithering. . . . . . . . . . . . . . . . . . . . 47 down.exe . . . . . . . . . . . . . . . . . . . . 52 field of view angle. . . . . . . . . . . . . . . 17 file extensions. . . . . . . . . . . . . . . . . 5 file formats . . . . . . . . . . . . . . . . . . 53 fixed palettes . . . . . . . . . . . . . . . . . 47 Floyd Steinberg dithering. . . . . . . . . . . . 47 focal_length . . . . . . . . . . . . . . . . . . 21 from . . . . . . . . . . . . . . . . . . . . . . 17 fuzz . . . . . . . . . . . . . . . . . . . . . . 34 gif. . . . . . . . . . . . . . . . . . . . . . . 46 global clipping. . . . . . . . . . . . . . . . . 45 haze . . . . . . . . . . . . . . . . . . . . . . 19 img file format. . . . . . . . . . . . . . . . . 53 img2gif.exe. . . . . . . . . . . . . . . . . . . 46 include files. . . . . . . . . . . . . . . . . . 5 input file format. . . . . . . . . . . . . . . . 13 jitter . . . . . . . . . . . . . . . . . . . . . 21 lights . . . . . . . . . . . . . . . . . . . . . 24 directional . . . . . . . . . . . . . . . 24 point . . . . . . . . . . . . . . . . . . 24 spherical . . . . . . . . . . . . . . . . 25 spot. . . . . . . . . . . . . . . . . . . 25 macro. . . . . . . . . . . . . . . . . . . . . . 15 macro continuation . . . . . . . . . . . . . . . 15 55 mandelbrot . . . . . . . . . . . . . . . . . . . 36 map file format. . . . . . . . . . . . . . . . . 53 median cut algorithm . . . . . . . . . . . . . . 46 noise. . . . . . . . . . . . . . . . . . . . . . 34 no_antialias . . . . . . . . . . . . . . . . . . 29 no_exp_trans . . . . . . . . . . . . . . . . . . 21 no_shadows . . . . . . . . . . . . . . . . . . . 21 ordered dither . . . . . . . . . . . . . . . . . 48 palette file . . . . . . . . . . . . . . . . . . 46 paste.exe. . . . . . . . . . . . . . . . . . . . 50 polygon. . . . . . . . . . . . . . . . . . . . . 39 popularity algorithm . . . . . . . . . . . . . . 46 preprocessor . . . . . . . . . . . . . . . . . . 15 projection . . . . . . . . . . . . . . . . . . . 18 random noise dithering . . . . . . . . . . . . . 48 resolution . . . . . . . . . . . . . . . . . . . 18 resume . . . . . . . . . . . . . . . . . . . . . 6 right-handed coordinate system . . . . . . . . . 13 ring . . . . . . . . . . . . . . . . . . . . . . 38 samples. . . . . . . . . . . . . . . . . . . . . 21 solid texture. . . . . . . . . . . . . . . . . . 33 sphere . . . . . . . . . . . . . . . . . . . . . 37 spherical pattern. . . . . . . . . . . . . . . . 33 start. . . . . . . . . . . . . . . . . . . . . . 18 statistics . . . . . . . . . . . . . . . . . . . 7 stop . . . . . . . . . . . . . . . . . . . . . . 18 studio . . . . . . . . . . . . . . . . . . . . . 17 surfaces . . . . . . . . . . . . . . . . . . . . 28 threshold. . . . . . . . . . . . . . . . . . . . 20 transform. . . . . . . . . . . . . . . . . . . . 42 transform_pop. . . . . . . . . . . . . . . . . . 42 triangular patch . . . . . . . . . . . . . . . . 40 turbulence . . . . . . . . . . . . . . . . . 32, 34 up . . . . . . . . . . . . . . . . . . . . . . . 17 up.exe . . . . . . . . . . . . . . . . . . . . . 51 viewpoint. . . . . . . . . . . . . . . . . . . . 17 wave bump map. . . . . . . . . . . . . . . . . . 31 xyzzy.v. . . . . . . . . . . . . . . . . . . . 5, 15