Comments on text of VP9 Bitstream & Decoding Process Specification v0.6

[Albert Liu]

Reading the VP9 spec went smoothly for the most part, however I had some comments:

• 3 Symbols (and abbreviated terms): SEG_LVL_*, DCT/ADST options, and reset_frame_context options could be expressed as enums.
  
• 3 Symbols (and abbreviated terms): Perhaps add symbols for the number of frame contexts, maximum number of subblocks per mode info, number of planes, and number of token types.
  
• 6.1 Frame syntax: Perhaps header_size_in_bytes could be renamed to compressed_header_bytes to avoid confusion with headerBytes.
  
• 6.2 Uncompressed header syntax: Many of the loop bounds in this call tree are expressed as integer literals rather than the corresponding symbols.
  
• 6.4.24 Token syntax: The variable Tokens is really the coefs derived from the tokens rather than the tokens themselves.
• 7.4.16 Token semantics: The names DCT_VAL_CAT1 etc. are used here, but in 9.3.1 Tree selection process, they are called DCT_VAL_CATEGORY1 etc.
• 9.3.2 Probability selection process: Under "The probability depends on the value of FrameIsIntra" the two cases are reversed. I.e. If FrameIsIntra is equal to 0, the non-keyframe probabilities should be used.
  
• B.2.1 Superframe index: frame_sizes[ i ] is specified as having type f(SzBytes). However, f() is supposed to take in a number of bits, not a number of bytes. Furthermore, it appears that the bytes are supposed to be stored in little-endian order rather than the big-endian order implied by f().