Discovered On2 Codec Headers
by Alex Beregszaszi
November 25, 2003

Contents
--------
 * Duck / TrueMotion (1)
 * TrueMotion 2
 * VP3.0
 * VP3.0b and up (3.1 and 3.2)
 * VP4 and VP5
 * VP6
 * Theora <3.3.0
 * Theora >=3.3.0


Duck / TrueMotion (1)
---------------------

byte 0: size of the header
byte 1: compression type (?)
byte 2: delta table indicator
byte 3: table (?)
bytes 4-5: height
bytes 6-7: width

The header is scrambled, here's a pseudo code for decoding:

real_header_size = ((header_size >> 5) || (header_size << 3)) & 0x7f

for (0 < i < real_header_size)
    header[i-1] = inbuf[i] ^ inbuf[i+1]

bytes 8-9: checksum
byte 10: version
byte 11: header type

Header types:
1: as above
2: 1 byte: frameinfo
   1 byte: control

TrueMotion 2
------------

bytes 0-3: type (two known types: 0x000000100 (~type1) and 0x000000101 (~type2))
bytes 4-7: format size (?)

If type1 stream, additional headers look like:

bytes 8-11: data size
bytes 12-13: height
bytes 14-15: width
bytes 16-19: compression flags
byte 20: unknown (it's in relation to height)
byte 21: unknown (it's in relation to width)

If type2 stream, additional headers look like:

bytes 8-11: header size (?)
bytes 12-15: data size
bytes 16-19: height
bytes 20-23: width
bytes 24-27: compression flags
bytes 28-31: unknown (it's in relation to height)
bytes 32-35: unknown (it's in relation to width)

Bit zero of compression flags indicates a file encoded with a shareware encoder,
so displaying the On2 logo is forced :)


VP3.0
-----

bit 7: 0 = intra frame, 1 = inter frame
bit 6: unused
bits 5-0: Quality index (0..63)

Additional headers come if frame is intra-coded:

bits 7-4: width code
bits 4-0: height code


VP3.0b and up (3.1 and 3.2)
---------------------------

bit 7: 0 = intra frame, 1 = inter frame
bit 6: unused
bits 5-0: Quality index (0..63)

Additional headers come if frame is intra-codec:

byte 0:
bits 7-4: width code (always zero)
bits 4-0: height code (always zero)

byte 1:
bits 7-3: version number (1 for vp3.1, 2 for vp3.2 (?))
bit 2: keyframe coding method (0 = dct)
bits 1-0: unused


VP4 and VP5
-----------

The same as VP3.0b.

Version codes:
3: VP4.0
4: ? (maybe VP4.1)
5: VP5.0


VP6
---

Header structure isn't discovered yet, but it might be based on some H.264 draft.


Theora <3.3.0
-------------

1 bit: frame indicator
1 bit: 0 = intra frame, 1 = inter frame
6 bits: Quality index (0..63)

Additional headers come if frame is intra-codec:

1 bit: keyframe coding method (0 = dct)
2 bits: unused


Theora >=3.3.0
--------------

1 bit: frame indicator
1 bit: 0 = intra frame, 1 = inter frame
6 bits: Quality index (0..63)
1 bit: unused

Additional headers come if frame is intra-codec:

1 bit: keyframe coding method (0 = dct)
2 bits: unused

In case of zero frame indicator, additional headers follow:

7 bit: header type

Header types:
 0: main header
 1: comments
 2: tables

Mean header:
 8 bits: version major
 8 bits: version minor
 8 bits: version micro
 16 bits: encoded width / 4
 16 bits: encoded height / 4
 24 bits: decoded width
 24 bits: decoded height
 8 bits: offset x (?)
 8 bits: offset y (?)
 32 bits: fps numerator
 32 bits: fps denumerator
 24 bits: aspect numerator
 24 bits: aspect denumerator
 
 If Theora <3.3.0:
  5 bits: keyframe frequency force
 
 8 bits: colorspace
 24 bits: bitrate
 6 bits: last (?) quality index
 
 If Theora >=3.3.0:
  5 bits: keyframe frequency force
  5 bits: unused

 Colorspaces:
 0: unspecified
 1: ITU 601
 2: CIE 709

Comments:
 32 bits: vendor len
 data: vendor_len bytes string without trailing zero
 
 32 bits: number of comments
 
 comment entries:
  32 bits: string len
  data: string_len bytes string without trailing zero

Tables:
 64*16 bits: quality threshold
 64*16 bits: dc scale factor
 64*8 bits: intraframe luma dequantizers
 64*8 bits: intraframe chroma dequatizers
 64*8 bits: interframe dequantizers


References
----------

FFmpeg sources:
http://www.ffmpeg.org/

Theora sources:
http://www.theora.org/