MPEG-4 features

The following features are supported for MPEG-4:

Info

The "info" feature for MPEG-4 exports some information about the picture type and the macroblock types. This feature is purely informative, no changes here will be applied back when transplicating.

TODO: string:
Picture type. This can either be:

  • "I" for I-frames
  • "P" for P-frames
  • "B" for B-frames
  • "S" for S(GMC)-frames

TODO: boolean:

  • true for interlaced frame (note: I have never seen a file that has this set to true)
  • false for non-interlaced frame

TODO: string: (optional when interlaced is true)

  • top for top field.
  • bottom for bottom field;

TODO:

  • null if the macroblock was skipped
  • a string, where each letter in the string corresponds to a certain macroblock feature:
    • 'I' for an intra macroblock (I, P, and S(GMC) frames)
    • 'a' for macroblocks with AC Prediction (I frames)
    • 'q' for macroblocks that change the quantization scale (I, P, B, and S(GMC) frames)
    • 'f' for forward motion vectors (P, B, and S(GMC) frames)
    • 'b' for backward motion vectors (B frames)
    • 'd' for direct motion vectors (B frames)
    • 'G' for Global Motion Compensation (S(GMC) frames)
    • '4' for macroblocks with 4 8x8 motion vectors instead of 1 16x16 (P and S(GMC) frames)
    • 'i' for Interlaced macroblocks (motion vectors are 16x8 in this case) (P, B, and S(GMC) frames)
    • '1' for macroblocks that change the DCT coefficients of the first luma block
    • '2' for macroblocks that change the DCT coefficients of the second luma block
    • '3' for macroblocks that change the DCT coefficients of the third luma block
    • '4' for macroblocks that change the DCT coefficients of the fourth luma block
    • '5' for macroblocks that change the DCT coefficients of the U chroma block
    • '6' for macroblocks that change the DCT coefficients of the V chroma block

Motion vectors

The "mv" feature for MPEG-4 exports the motion vectors that are encoded in the bitstream.

There may be "forward" and "backward" motion vectors.

TODO: MV:

  • null if there is no motion vector delta for this macroblock
  • an Array with two numbers, the first being horizontal and the second being vertical.
  • an Array with 4 elements (in the case where there are 4 motion vectors per macroblock). Each element is either null or an Array with two numbers, the first being horizontal and the second being vertical.

TODO: (informative) fcode:

  • fcode is an Array with two numbers, the first being horizontal and the second being vertical.
  • This field is informative. It tells you the fcode that was encoded for this frame. This can be used to calculate the minimum and maximum values allowed for the motion vectors.
  • For MPEG-4, the motion vector range is calculated by: 
    const min_val = -(1<<(4 + fcode));
const max_val =  (1<<(4 + fcode))-1;

TODO: string:

  • The "overflow" field will instruct FFglitch on what to do when it imports motion vector values that overflow the limit defined by fcode. Possible values are:
    • "assert": exits FFglitch on overflow.
    • "truncate": truncates motion vector values within range (I suggest using this one).
    • "ignore": does nothing (allows overflow).
    • "warn": warns once, but then ignores.

Motion vectors (delta only)

The "mv_delta" feature for MPEG-4 is similar to the "mv" feature, but the motion vectors are expressed as a delta value from the previous motion vector.

There may be "forward" and "backward" motion vectors.

TODO: MV_delta:

  • null if there is no motion vector delta for this macroblock
  • an Array with two numbers, the first being horizontal and the second being vertical.
  • an Array with 4 elements (in the case where there are 4 motion vectors per macroblock). Each element is either null or an Array with two numbers, the first being horizontal and the second being vertical.

TODO: (informative) fcode:

  • fcode is an Array with two numbers, the first being horizontal and the second being vertical.
  • This field is informative. It tells you the fcode that was encoded for this frame. This can be used to calculate the minimum and maximum values allowed for the motion vectors.
  • For MPEG-4, the motion vector range is calculated by: 
    const min_val = -(1<<(4 + fcode));
const max_val =  (1<<(4 + fcode))-1;

TODO: string:

  • The "overflow" field will instruct FFglitch on what to do when it imports motion vector values that overflow the limit defined by fcode. Possible values are:
    • "assert": exits FFglitch on overflow.
    • "truncate": truncates motion vector values within range (I suggest using this one).
    • "ignore": does nothing (allows overflow).
    • "warn": warns once, but then ignores.

Macroblock

The "mb" feature for MPEG-4 exports the bytestream of each macroblock as a string value.

Macroblocks are mostly self-sufficient, so it might be possible to reorder them and still have a valid bitstream.

TODO: macroblock:

  • null if the macroblock was skipped.
  • a string with the hexadecimal representation of the bytestream for this macroblock

TODO: (informative) number:

  • the size in bits of the macroblock. This is used by FFglitch to properly align the bitstream when transplicating. Do not change these values.

Global motion compensation

The "gmc" feature for MPEG-4 exports the Global Motion Compensation parameters for a frame.

gmc is an Array with up to 3 elements, each being an Array with 2 numbers each.

I don’t remember how the gmc calculation is done. I’ll document this later if anyone reeeeally wants it.