MD2 (file Format) - File Format

File Format

An MD2 file begins with a fixed length header followed by static model data such as texture coordinates. Dynamic data such as vertices and normals are stored within a number of file chunks called frames (or key-frames) which each have their own short headers.

In defining the file structure several data types will be referred to.

int (4 bytes), short (2 bytes), and char (1 byte)

MD2 Header

Offset	Data type	Name	Description
0	int	ident	Magic number. Must be equal to "IDP2"
4	int	version	MD2 version. Must be equal to 8
8	int	skinwidth	Width of the texture
12	int	skinheight	Height of the texture
16	int	framesize	Size of one frame in bytes
20	int	num_skins	Number of textures
24	int	num_xyz	Number of vertices
28	int	num_st	Number of texture coordinates
32	int	num_tris	Number of triangles
36	int	num_glcmds	Number of OpenGL commands
40	int	num_frames	Total number of frames
44	int	ofs_skins	Offset to skin names (each skin name is an unsigned char and are null terminated)
48	int	ofs_st	Offset to s-t texture coordinates
52	int	ofs_tris	Offset to triangles
56	int	ofs_frames	Offset to frame data
60	int	ofs_glcmds	Offset to OpenGL commands
64	int	ofs_end	Offset to end of file

At the offset ofs_st there are num_st of this structure:

Data type	Name
short	s
short	t

To recover the floating-point texture coordinates as used by common 3D display API's such as OpenGL, divide the texture coordinates by the respective size dimensions of the texture:

sfloat = (float)s / texturewidth tfloat = (float)t / textureheight

At offset ofs_tris there are num_tris of the following structure

short vertexindex short textureindex

These are indices to the vertices and texture coordinates.

At offset ofs_frames frame data is stored, each frame has a short header followed by a number of vertex and surface normal indices, the frame header structure is like this:

float scale float translate char name

Then there is num_xyz of this structure:

unsigned char v unsigned char lightnormalindex

Each vertex is stored as an integer array. To recover the floating-point vertex coordinates, the MD2 reader multiplies each coordinate by the scaling vector for the current frame and then adds the frame's translation vector; these vectors can be found in the frame's header. Alternatively, you can set the translation/scale matrix with model's values before rendering the model's frame (e.g. if using OpenGL); please note that scaling using glScale may de-normalize the surface normals.