.M2V File Format – Macroblocks

Macroblock syntax

The first macroblock of a slice has address mb_row × mb_width − 1 plus the increment; a jump of more than one skips macroblocks. I pictures may not skip, except in spatial or SNR scalable enhancement layers; a B picture may not skip straight after an intra macroblock; and the first and last macroblocks of a slice are never skipped. macroblock_stuffing from MPEG-1 is not allowed.

Macroblock address increment

Table B.1.

Macroblock types

The table depends on the picture type and the scalable mode. Each code sets six flags: Q macroblock_quant, F macroblock_motion_forward, B macroblock_motion_backward, P macroblock_pattern, I macroblock_intra and W spatial_temporal_weight_code_flag, and allows certain spatial-temporal weight classes (class 0 temporal prediction only, 4 spatial only, 1 to 3 a mix; clause 7.7.4).

I pictures (Table B.2)

P pictures (Table B.3)

B pictures (Table B.4)

I pictures with spatial scalability (Table B.5)

P pictures with spatial scalability (Table B.6)

B pictures with spatial scalability (Table B.7)

SNR scalability, all picture types (Table B.8)

The SNR enhancement layer needs a “Not Coded” type because the first and last macroblocks of a slice cannot be skipped.

Motion types and DCT type

Tables 6-17 and 6-18. When frame_pred_frame_dct is 1 the frame motion type is absent and frame-based prediction is used. Intra macroblocks with concealment vectors have no motion type and use frame-based (frame pictures) or field-based (field pictures) vectors. When dct_type is absent it is unused in field pictures and uncoded or skipped macroblocks, and 0 (frame DCT) when frame_pred_frame_dct is 1 (Table 6-19).

Quantiser scale

Table 7-6. q_scale_type in the picture coding extension picks the column. Code 0 is forbidden.

Motion vectors

With one vector, motion_vertical_field_select[0][s] (1 bit: 0 top reference field, 1 bottom) comes first only for field-format vectors that are not dual prime; with two vectors each has its own select bit. Each vector is:

Indices (Table 7-7): r is the first (0) or second (1) vector, s forward (0) or backward (1), t horizontal (0) or vertical (1).

Range of a decoded vector component in half samples, by f_code (Table 7-8). The first column applies to the vertical component of field vectors in frame pictures.

The levels cap f_code: horizontal 1 to 7 at Low level, 1 to 8 at Main and 1 to 9 above; vertical 1 to 4, 5, 5, 5 in frame pictures and 1 to 3, 4, 4, 4 in field pictures (Table 8-8).

Coded block pattern

Table B.9 gives cbp, a 6-bit pattern for blocks 0 to 5 (bit 5 is block 0, the top-left luma block; blocks 4 and 5 are Cb and Cr). For 4:2:2 the 2 extra bits cover blocks 6 and 7, and for 4:4:4 the 6 extra bits cover blocks 6 to 11, again most significant bit first. Luma blocks are 0 to 3; in 4:2:2 blocks 4 and 6 are Cb and 5 and 7 Cr; in 4:4:4 the even-numbered blocks from 4 are Cb and the odd ones Cr (Table 7-1). In intra macroblocks every block is coded.

Block syntax

The DC differential is added to a predictor per colour component, which is reset at the start of each slice, at each non-intra macroblock and at each skip, to 128, 256, 512 or 1024 for intra_dc_precision 0 to 3 (Table 7-2). Coefficients are then placed by the scan order and inverse quantised. Data partitioning changes the block layer; see clause 7.10.

DCT coefficients

Each code gives a run of zero coefficients and a level; the final bit s is the sign of the level (0 positive, 1 negative). Table B.14 (table zero) is used for non-intra blocks, and for intra blocks when intra_vlc_format is 0; Table B.15 (table one) for intra blocks when it is 1 (Table 7-3). In table zero, run 0 level 1 is 1s for the first coefficient of a non-intra block and 11s otherwise, so end of block 10 cannot be the first code. An intra block’s AC coefficients start after the DC value.

After the escape code come 6 bits of run (0 to 63) and 12 bits of signed level in two’s complement: −2047 to +2047, with 0 forbidden and 1000 0000 0000 reserved (Table B.16). This escape differs from MPEG-1’s.

The longer codes, from 13 to 17 bits including the sign, are the same in both tables, except that the first four (run 0, levels 12 to 15) exist only in table zero; table one codes those with 1111 1010 s to 1111 1111 s above:

Scan orders

Coefficients are read in one of two orders, chosen by alternate_scan. Each grid shows, at each position of the 8 × 8 block (row v, column u), the index of the coefficient that lands there (Figures 7-2 and 7-3). Quantiser matrices are always sent in the zig-zag order.

zig-zag (alternate_scan 0)
 0  1  5  6 14 15 27 28
 2  4  7 13 16 26 29 42
 3  8 12 17 25 30 41 43
 9 11 18 24 31 40 44 53
10 19 23 32 39 45 52 54
20 22 33 38 46 51 55 60
21 34 37 47 50 56 59 61
35 36 48 49 57 58 62 63

alternate (alternate_scan 1)
 0  4  6 20 22 36 38 52
 1  5  7 21 23 37 39 53
 2  8 19 24 34 40 50 54
 3  9 18 25 35 41 51 55
10 17 26 30 42 46 56 60
11 16 27 31 43 47 57 61
12 15 28 32 44 48 58 62
13 14 29 33 45 49 59 63

Data partitioning

With scalable_mode 0, each slice of partition 0 carries a priority_breakpoint that says where the data moves to partition 1 (Table 7-30):

Partition 1 repeats the sequence, GOP and picture headers and the sequence end code; its only extensions are the sequence, picture coding and sequence scalable extensions.