World of Video

home *** CD-ROM | disk | FTP | other *** search

/ World of Video / World of Video.iso / gfxprograms / viewers / viewtek / anim7.doc next >

Wrap

Text File | 1995-02-13 | 29KB | 602 lines

A N I M An IFF Format For CEL Animations Revision date: 4 May 1988 prepared by: SPARTA Inc. 23041 de la Carlota Laguna Hills, Calif 92653 (714) 768-8161 contact: Gary Bonham also by: Aegis Development Co. 2115 Pico Blvd. Santa Monica, Calif 90405 213) 392-9972 Anim7 Appendix (july 92) by: Wolfgang Hofer A-2722 Winzendorf Wr. Neustaedterstr. 140 1.0 Introduction The ANIM IFF format was developed at Sparta originally for the production of animated video sequences on the Amiga computer. The intent was to be able to store, and play back, sequences of frames and to minimize both the storage space on disk (through compression) and playback time (through efficient de-compression algorithms). It was desired to maintain maximum compatibility with existing IFF formats and to be able to display the initial frame as a normal still IFF picture. Several compression schemes have been introduced in the ANIM format. Most of these are strictly of historical interest as the only one currently being placed in new code is the vertical run length encoded byte encoding developed by Jim Kent. 1.1 ANIM Format Overview The general philosophy of ANIMs is to present the initial frame as a normal, run-length-encoded, IFF picture. Subsequent frames are then described by listing only their differences from a previous frame. Normally, the "previous" frame is two frames back as that is the frame remaining in the hidden screen buffer when double-buffering is used. To better understand this, suppose one has two screens, called A and B, and the ability to instantly switch the display from one to the other. The normal playback mode is to load the initial frame into A and duplicate it into B. Then frame A is displayed on the screen. Then the differences for frame 2 are used to alter screen B and it is displayed. Then the differences for frame 3 are used to alter screen A and it is displayed, and so on. Note that frame 2 is stored as differences from frame 1, but all other frames are stored as differences from two frames back. ANIM is an IFF FORM and its basic format is as follows (this assumes the reader has a basic understanding of IFF format files): FORM ANIM . FORM ILBM first frame . . BMHD normal type IFF data . . ANHD optional animation header chunk for timing of 1st frame. . . CMAP . . BODY . FORM ILBM frame 2 . . ANHD animation header chunk . . DLTA delta mode data . FORM ILBM frame 3 . . ANHD . . DLTA ... The initial FORM ILBM can contain all the normal ILBM chunks, such as CRNG, etc. The BODY will normally be a standard run-length-encoded data chunk (but may be any other legal compression mode as indicated by the BMHD). If desired, an ANHD chunk can appear here to provide timing data for the first frame. If it is here, the operation field should be =0. The subsequent FORMs ILBM contain an ANHD, instead of a BMHD, which duplicates some of BMHD and has additional parameters pertaining to the animation frame. The DLTA chunk contains the data for the delta compression modes. If the older XOR compression mode is used, then a BODY chunk will be here. In addition, other chunks may be placed in each of these as deemed necessary (and as code is placed in player programs to utilize them). A good example would be CMAP chunks to alter the color palette. A basic assumption in ANIMs is that the size of the bitmap, and the display mode (e.g. HAM) will not change through the animation. Take care when playing an ANIM that if a CMAP occurs with a frame, then the change must be applied to both buffers. Note that the DLTA chunks are not interleaved bitmap representations, thus the use of the ILBM form is inappropriate for these frames. However, this inconsistency was not noted until there were a number of commercial products either released or close to release which generated/played this format. Therefore, this is probably an inconsistency which will have to stay with us. 1.2 Recording ANIMs To record an ANIM will require three bitmaps - one for creation of the next frame, and two more for a "history" of the previous two frames for performing the compression calculations (e.g. the delta mode calculations). There are five frame-to-frame compression methods currently defined. The first three are mainly for historical interest. The product Aegis VideoScape 3D utilizes the third method in version 1.0, but switched to method 5 on 2.0. This is the only instance known of a commercial product generating ANIMs of any of the first three methods. The fourth method is a general short or long word compression scheme which has several options including whether the compression is horizontal or vertical, and whether or not it is XOR format. This offers a choice to the user for the optimization of file size and/or playback speed. The fifth method is the byte vertical run length encoding as designed by Jim Kent. Do not confuse this with Jim's RIFF file format which is different than ANIM. Here we utilized his compression/decompression routines within the ANIM file structure. The following paragraphs give a general outline of each of the methods of compression currently included in this spec. 1.2.1 XOR mode This mode is the original and is included here for historical interest. In general, the delta modes are far superior. The creation of XOR mode is quite simple. One simply performs an exclusive-or (XOR) between all corresponding bytes of the new frame and two frames back. This results in a new bitmap with 0 bits wherever the two frames were identical, and 1 bits where they are different. Then this new bitmap is saved using run-length-encoding. A major obstacle of this mode is in the time consumed in performing the XOR upon reconstructing the image. 1.2.2 Long Delta mode This mode stores the actual new frame long-words which are different, along with the offset in the bitmap. The exact format is shown and discussed in section 2 below. Each plane is handled separately, with no data being saved if no changes take place in a given plane. Strings of 2 or more long-words in a row which change can be run together so offsets do not have to be saved for each one. Constructing this data chunk usually consists of having a buffer to hold the data, and calculating the data as one compares the new frame, long-word by long-word, with two frames back. 1.2.3 Short Delta mode This mode is identical to the Long Delta mode except that short-words are saved instead of long-words. In most instances, this mode results in a smaller DLTA chunk. The Long Delta mode is mainly of interest in improving the playback speed when used on a 32-bit 68020 Turbo Amiga. 1.2.4 General Delta mode The above two delta compression modes were hastily put together. This mode was an attempt to provide a well-thought-out delta compression scheme. Options provide for both short and long word compression, either vertical or horizontal compression, XOR mode (which permits reverse playback), etc. About the time this was being finalized, the fifth mode, below, was developed by Jim Kent. In practice the short-vertical-run-length-encoded deltas in this mode play back faster than the fifth mode (which is in essence a byte-vertical-run-length-encoded delta mode) but does not compress as well - especially for very noisy data such as digitized images. In most cases, playback speed not being terrifically slower, the better compression (sometimes 2x) is preferable due to limited storage media in most machines. Details on this method are contained in section 2.2.2 below. 1.2.5 Byte Vertical Compression This method does not offer the many options that method 4 offers, but is very successful at producing decent compression even for very noisy data such as digitized images. The method was devised by Jim Kent and is utilized in his RIFF file format which is different than the ANIM format. The description of this method in this document is taken from Jim's writings. Further, he has released both compression and decompression code to public domain. Details on this method are contained in section 2.2.3 below. 1.3 Playing ANIMs Playback of ANIMs will usually require two buffers, as mentioned above, and double-buffering between them. The frame data from the ANIM file is used to modify the hidden frame to the next frame to be shown. When using the XOR mode, the usual run- length-decoding routine can be easily modified to do the exclusive-or operation required. Note that runs of zero bytes, which will be very common, can be ignored, as an exclusive or of any byte value to a byte of zero will not alter the original byte value. The general procedure, for all compression techniques, is to first decode the initial ILBM picture into the hidden buffer and double- buffer it into view. Then this picture is copied to the other (now hidden) buffer. At this point each frame is displayed with the same procedure. The next frame is formed in the hidden buffer by applying the DLTA data (or the XOR data from the BODY chunk in the case of the first XOR method) and the new frame is double-buffered into view. This process continues to the end of the file. A master colormap should be kept for the entire ANIM which would be initially set from the CMAP chunk in the initial ILBM. This colormap should be used for each frame. If a CMAP chunk appears in one of the frames, then this master colormap is updated and the new colormap applies to all frames until the occurrance of another CMAP chunk. Looping ANIMs may be constructed by simply making the last two frames identical to the first two. Since the first two frames are special cases (the first being a normal ILBM and the second being a delta from the first) one can continually loop the anim by repeating from frame three. In this case the delta for creating frame three will modify the next to the last frame which is in the hidden buffer (which is identical to the first frame), and the delta for creating frame four will modify the last frame which is identical to the second frame. Multi-File ANIMs are also supported so long as the first two frames of a subsequent file are identical to the last two frames of the preceeding file. Upon reading subsequent files, the ILBMs for the first two frames are simply ignored, and the remaining frames are simply appended to the preceeding frames. This permits splitting ANIMs across multiple floppies and also permits playing each section independently and/or editing it independent of the rest of the ANIM. Timing of ANIM playback is easily achieved using the vertical blank interrupt of the Amiga. There is an example of setting up such a timer in the ROM Kernel Manual. Be sure to remember the timer value when a frame is flipped up, so the next frame can be flipped up relative to that time. This will make the playback independent of how long it takes to decompress a frame (so long as there is enough time between frames to accomplish this decompression). 2.0 Chunk Formats 2.1 ANHD Chunk The ANHD chunk consists of the following data structure: UBYTE operation The compression method: =0 set directly (normal ILBM BODY), =1 XOR ILBM mode, =2 Long Delta mode, =3 Short Delta mode, =4 Generalized short/long Delta mode, =5 Byte Vertical Delta mode =7 short/long Vertical Delta mode =74 (ascii 'J') reserved for Eric Graham's compression technique (details to be released later). UBYTE mask (XOR mode only - plane mask where each bit is set =1 if there is data and =0 if not.) UWORD w,h (XOR mode only - width and height of the area represented by the BODY to eliminate unnecessary un-changed data) WORD x,y (XOR mode only - position of rectangular area representd by the BODY) ULONG abstime (currently unused - timing for a frame relative to the time the first frame was displayed - in jiffies (1/60 sec)) ULONG reltime (timing for frame relative to time previous frame was displayed - in jiffies (1/60 sec)) UBYTE interleave (unused so far - indicates how may frames back this data is to modify. =0 defaults to indicate two frames back (for double buffering). =n indicates n frames back. The main intent here is to allow values of =1 for special applications where frame data would modify the immediately previous frame) UBYTE pad0 Pad byte, not used at present. ULONG bits 32 option bits used by options=4 and 5. At present only 6 are identified, but the rest are set =0 so they can be used to implement future ideas. These are defined for option 4 only at this point. It is recommended that all bits be set =0 for option 5 and that any bit settings used in the future (such as for XOR mode) be compatible with the option 4 bit settings. Player code should check undefined bits in options 4 and 5 to assure they are zero. The six bits for current use are: bit # set =0 set =1 =============================================== 0 short data long data 1 set XOR 2 separate info one info list for each plane for all planes 3 not RLC RLC (run length coded) 4 horizontal vertical 5 short info offsets long info offsets UBYTE pad[16] This is a pad for future use for future compression modes. 2.2 DLTA Chunk This chunk is the basic data chunk used to hold delta compression data. The format of the data will be dependent upon the exact compression format selected. At present there are two basic formats for the overall structure of this chunk. 2.2.1 Format for methods 2 & 3 This chunk is a basic data chunk used to hold the delta compression data. The minimum size of this chunk is 32 bytes as the first 8 long-words are byte pointers into the chunk for the data for each of up to 8 bitplanes. The pointer for the plane data starting immediately following these 8 pointers will have a value of 32 as the data starts in the 33-rd byte of the chunk (index value of 32 due to zero-base indexing). The data for a given plane consists of groups of data words. In Long Delta mode, these groups consist of both short and long words - short words for offsets and numbers, and long words for the actual data. In Short Delta mode, the groups are identical except data words are also shorts so all data is short words. Each group consists of a starting word which is an offset. If the offset is positive then it indicates the increment in long or short words (whichever is appropriate) through the bitplane. In other words, if you were reconstructing the plane, you would start a pointer (to shorts or longs depending on the mode) to point to the first word of the bitplane. Then the offset would be added to it and the following data word would be placed at that position. Then the next offset would be added to the pointer and the following data word would be placed at that position. And so on... The data terminates with an offset equal to 0xFFFF. A second interpretation is given if the offset is negative. In that case, the absolute value is the offset+2. Then the following short-word indicates the number of data words that follow. Following that is the indicated number of contiguous data words (longs or shorts depending on mode) which are to be placed in contiguous locations of the bitplane. If there are no changed words in a given plane, then the pointer in the first 32 bytes of the chunk is =0. 2.2.2 Format for method 4 The DLTA chunk is modified slightly to have 16 long pointers at the start. The first 8 are as before - pointers to the start of the data for each of the bitplanes (up to a theoretical max of 8 planes). The next 8 are pointers to the start of the offset/numbers data list. If there is only one list of offset/numbers for all planes, then the pointer to that list is repeated in all positions so the playback code need not even be aware of it. In fact, one could get fancy and have some bitplanes share lists while others have different lists, or no lists (the problems in these schemes lie in the generation, not in the playback). The best way to show the use of this format is in a sample playback routine. SetDLTAshort(bm,deltaword) struct BitMap *bm; WORD *deltaword; { int i; LONG *deltadata; WORD *ptr,*planeptr; register int s,size,nw; register WORD *data,*dest; deltadata = (LONG *)deltaword; nw = bm->BytesPerRow >>1; for (i=0;i<bm->Depth;i++) { planeptr = (WORD *)(bm->Planes[i]); data = deltaword + deltadata[i]; ptr = deltaword + deltadata[i+8]; while (*ptr != 0xFFFF) { dest = planeptr + *ptr++; size = *ptr++; if (size < 0) { for (s=size;s<0;s++) { *dest = *data; dest += nw; } data++; } else { for (s=0;s<size;s++) { *dest = *data++; dest += nw; } } } } return(0); } The above routine is for short word vertical compression with run length compression. The most efficient way to support the various options is to replicate this routine and make alterations for, say, long word or XOR. The variable nw indicates the number of words to skip to go down the vertical column. This one routine could easily handle horizontal compression by simply setting nw=1. For ultimate playback speed, the core, at least, of this routine should be coded in assembly language. 2.2.2 Format for method 5 In this method the same 16 pointers are used as in option 4. The first 8 are pointers to the data for up to 8 planes. The second set of 8 are not used but were retained for several reasons. First to be somewhat compatible with code for option 4 (although this has not proven to be of any benefit) and second, to allow extending the format for more bitplanes (code has been written for up to 12 planes). Compression/decompression is performed on a plane-by-plane basis. For each plane, compression can be handled by the skip.c code (provided Public Domain by Jim Kent) and decompression can be handled by unvscomp.asm (also provided Public Domain by Jim Kent). Compression/decompression is performed on a plane-by-plane basis. The following description of the method is taken directly from Jim Kent's code with minor re-wording. Please refer to Jim's code (skip.c and unvscomp.asm) for more details: Each column of the bitplane is compressed separately. A 320x200 bitplane would have 40 columns of 200 bytes each. Each column starts with an op-count followed by a number of ops. If the op-count is zero, that's ok, it just means there's no change in this column from the last frame. The ops are of three classes, and followed by a varying amount of data depending on which class: 1. Skip ops - this is a byte with the hi bit clear that says how many rows to move the "dest" pointer forward, ie to skip. It is non-zero. 2. Uniq ops - this is a byte with the hi bit set. The hi bit is masked down and the remainder is a count of the number of bytes of data to copy literally. It's of course followed by the data to copy. 3. Same ops - this is a 0 byte followed by a count byte, followed by a byte value to repeat count times. Do bear in mind that the data is compressed vertically rather than horizontally, so to get to the next byte in the destination we add the number of bytes per row instead of one! ------------------------------------------------------------------------- Appendix for Anim7 Formats Wolfgang Hofer, 23.6.92 ------------------------------------------------------------------------- Anim method 7 is designed for maximum playback speed and acceptable packing rates (packing usually not as good as method 5, but more efficient than methodes 1 -- 4) Method 7 is not in the IFF specification today but supported by the Public Domain Programs AAP/AAC. #.# Chunk Sequence: Method 7 Anims should use the same Chunk Sequence as methods 1..5. Alternativley the first frame may have a DLTA chunk instead of the BODY chunk. In that case the DLTA is the difference to a 'black frame'. A player has to clear all bitplanes of the first bitmap to zero, and then call his DLTA unpack routines for this frame. FORM ANIM . FORM ILBM first frame . . BMHD normal type IFF data . . ANHD optional animation header chunk for timing of 1st frame. . . CMAP . . { BODY | full picture or DLTA } difference to 'black frame' . FORM ILBM frame 2 . . ANHD animation header chunk . . DLTA delta mode data . . [CMAP] . FORM ILBM frame 3 . . ANHD . . DLTA . . [CMAP] ... The initial FORM ILBM can contain all the normal ILBM chunks, such as CRNG, etc. The BODY will normally be a standard run-length-encoded data chunk (but may be any other legal compression mode as indicated by the BMHD). If desired, an ANHD chunk can appear here to provide timing data for the first frame. If it is here, the operation field should be =0. If the initial FORM ILBM uses a DLTA chunk, the ANHD chunk must appear, and the operation field must be set to the according anim method. # Chunk Formats #.# ANHD Chunk for method 7 The ANHD chunk consists of the following data structure: UBYTE operation The compression method: =7 short/long Vertical Delta mode UBYTE mask unused UWORD w,h unused WORD x,y unused ULONG abstime unused ULONG reltime (timing for frame relative to time previous frame was displayed - in jiffies (1/60 sec)) UBYTE interleave = 0 (see ANHD description above) UBYTE pad0 unused ULONG bits 32 option bits used by methode=4 and 5. methode 7 uses only bit #0 bit # set =0 set =1 =============================================== 0 short data long data UBYTE pad[16] unused #.# DLTA Chunk #.#.# Format for method 7 The DLTA Chunks of method7 consists of - 8 pointers to opcode lists - 8 pointers to data lists - data lists (long/short) - opcode lists (bytes) In this method the DLTA Chunk begins with 16 pointers. The first 8 longwords are pointers to the opcode lists for up to 8 planes. The second set of 8 longwords are pointers to the correspondig data lists. If there are less than 8 Planes all unused pointers are set to zero. Compression/decompression is performed on a plane-by-plane basis. The following description of the method is similar to Jim Kent's methode 5, except that data is stored in a seperated datalist (long or short, depending on bit#0 of the ANHD bits) and doesn't follow immediate after the opcode. In methode 7 the bitplane is splitted into vertical columns. Each column of the bitplane is compressed separately. A 320x200 bitplane would have 20 columns of 200 short datas each. (or 10 columns of 200 long datas) Each column starts with an op-count followed by a number of ops. If the op-count is zero, that's ok, it just means there's no change in this column from the last frame. The ops are of three classes. The ops refer to a varying amount of data (to fetch from the corresponding datalist) depending on which class: 1. Skip ops - this is a byte with the hi bit clear that says how many rows to move the "dest" pointer forward, ie to skip. It is non-zero. Skip ops have no corresponding data-items in the datalist. 2. Uniq ops - this is a byte with the hi bit set. The hi bit is masked down and the remainder is a count of the number of data to copy literally from the datalist to the "dest" pointer column. (Each data item to the next destination row) Data items may be long or short organized. 3. Same ops - this is a 0 byte followed by a count byte. The count byte says how many rows of the current column are to be set to the same data-item. the data-item (long or short) is fetched from the datalist. Do bear in mind that the data is compressed vertically rather than horizontally, so to get to the next address in the destination we have to add the number of bytes per row instead of 2 (or 4)!