In JCTVC-U1005-v2.doc, in 9.3.3.1 Table 9-42. Missing binarisation associations for:

tu_residual_act_flag palette_predictor_run num_signalled_palette_entries palette_transpose_flag

The following elements are repeated (no longer in transform_unit()):

cu_qp_delta_sign_flag cu_qp_delta_abs cu_chroma_qp_offset_idx cu_chroma_qp_offset_flag

The following elements have unused parameters in their binarisation process:

num_palette_indices_minus1, nCbS is unused palette_escape_val, qP is unused

In 9.3.4.2 Derivation process for ctxTable, ctxIdx and bypassFlag. Missing ctxIncs for:

palette_index_idc palette_run_refinement_bits palette_escape_val

Missing ctxTables for:

palette_index_idc (conditional on non-trivial ctxInc) palette_run_refinement_bits (conditional on non-trivial ctxInc) palette_escape_val (conditional on non-trivial ctxInc) palette_transpose_flag

In JCTVC-U1005-v2.doc, in 9.3.3.6 Truncated Binary (TB) binarization process. Equation block 9-17 says

u = 2k + 1 - n,

this should be

u = 2^{(k + 1)} - n, or u = 1 << (k + 1) - n.

Also the accompanying text doesn't make sense, in the invoked FL processes we want the variable fixedLength to equal k, k + 1 respectively. So consider introducing a binary binarisation process where we specify the number of bits with which we write the symbol value.

Then

– Otherwise, if synVal less than u, the TB bin string is derived by invoking the FL binarization process specified in clause 9.3.3.5 for synVal with the input symbolVal set to k and with cMax.

becomes

– Otherwise, if synVal less than u, the TB bin string is derived by invoking the Binary binarization process specified in clause X.X.X.X for synVal with numBits set to k.

And

– Otherwise (synVal greater than or equal to u), the TB bin string is derived by invoking the FL binarization process specified in clause 9.3.3.5 for synVal + u with the input symbolVal set to ( k + 1 ) and with cMax.

becomes

– Otherwise (synVal greater than or equal to u), the TB bin string is derived by invoking the Binary binarization process specified in clause X.X.X.X for synVal + u with numBits set to k + 1.

Finally consider replacing synVal with symbolVal for consistency with the other binarisation processes.

In JCTVC-U1005-v2.doc the specification for pred_weight_table has a test for the presence of luma_weight_l1_flag based on the value of RefPicList0[i].

I believe (based on examination of TDecCavlc::xParsePredWeightTable) that this test (and the one for chroma_weight_l1_flag) should instead be based on RefPicList1[i].

By the way, it feels really weird that the syntax decode of pred_weight_table (which is inside slice_segment_header) now has dependencies on RefPicList0 and PicOrderCntVal - variables that (according to the first sentence of 8.3.2) are only computed "after decoding of a slice header".