Rtm Binarised File Format

UnBinarised Format

For unbinarised see Rtm (Animation) File Format

Binarised format

Binarised rtm was introduced for the Arma2 engine.

Structure

RTMB
{
  RTMBHeader       Header;
  byte             Always1;              // ??
  XYZTriplet       Displacement;
  ulong            nFrames;              // 55 eg
  ulong            Generally0;           // ??
  ulong            nBones;               // same value as NoOfBones
  ulong            NoOfBones;            // 128 eg
  Asciiz           BoneNames[NoOfBones]; // "weapon\0\rightarm\leftarm\etc..."
  FrameTimes       FrameTimes;           // CompressedFloats
  FramePosition    FramePositions[nFrames];// CompressedFrames
}

BoneNames are the traditional Bis catenated strings. One after the other separated by '\0'. Unlike most others encountered in Bis file architecture, there is no final terminating double null because the count is already known.

RTMBHeader

RTMBHeader
{
 char[4]       Type;          //"BMTR" 
 ulong         Version;       // =3
}

FrameTimes (compressed)

FrameTimes
{
 ulong nFrameTimes;          // should always be same as nFrames
 float Array[nFrameTimes];
}

This is the traditional Bis potentially compressed structure, where, if, the total bytes stored >=1024, the array is compressed (in lzo format). All bones are 'held' in a position and orientation (specified by FramePosition) for the duration of that frametime period.

FramePosition (compressed)

FramePosition
{
 ulong   ThisFrameNoOfBones;               // Should always be same as nBones
 short   Transforms[ThisFrameNoOfBones][7];
}FramePositions[nFrames];

Each bone is held in this position and orientation for this frametime. ThisFrameNoOfBones is irrelevant to the engine and always must be the nBones value. It is present however because of efficiencies in using the standard bis lzo compression methods which rely on this 'count'.

The short values are compressed floats (7 in total) for each 'frame'. conversion to float is

float ShortToFloat(short value) {return ((float)1.0 / (65534)) * (value + 32767);}

this is suspiciously similar to ofp compressed floats

float = short * 0.045

4 floats probably are used to describe a quaternion that describes the rotation. the other 3 floats are then for the translation