JVET-AL0324 AHG9: A summary of proposals on the PRI SEI message [Y.-K. Wang (Bytedance)]

1. On the PRI SEI message being a layer-specific or cross-layer SEI message
   1. Since the persistence is currently specified in a layer-specific manner, remove pri_multilayer_flag, pri_region_layer_id[ i ], and pri_region_is_a_layer_flag[ i ] to make the entire design layer-specific. (JVET-AL0070 item 10)
   2. Make the interface variables for width, height, max width, and max height to be layer specific, in both VSEI and VVC. (JVET-AL0120 item 1, JVET-AL0129 item 10, JVET-AL0303 item 3)
   3. Specify the persistence of the PRI SEI message in a cross-layer manner. (JVET-AL0120 item 3, JVET-AL0303 item 2)
   4. Make the interface variables for chroma format, luma bit depth and chroma bit depth layer specific (JVET-AL0122 item 9, JVET-AL0303 item 5)

An alternative to d would be to constrain that they were all equal.

Adopt Item 1 b and c. Add the constraint that chroma format, luma bit depth and chroma bit depth layer are all equal.

1. Change pri_num_resampling_ratios_minus1 to pri_num_resampling_ratios. (JVET-AL0129 item 8)

No action was taken on this.

1. Change "i = 1" below to "i = 0": (JVET-AL0070 item 1)

No action was taken on this.

1. Modify the derivation process of resampling ratios to cover ratio=1 cases

No action was taken on this.

1. Specify the value range for pri_num_resampling_ratios_minus1 to be in the range of 0 to pri_num_regions_minus1, inclusive. (JVET-AL0120 item 5)

Adopt item 5.

1. (In JVET-AL_notes_d3: Further discussion requesed after checking if an inference value is needed) Add the following inference: When not present, the value of pri_resampling_ratio_idx[ i ] is inferred to be equal to 0. (JVET-AL0070 item 4 and JVET-AL0122 item 2)

Adopt Item 6.

1. Require uniqueness of resampling ratios by adding the following constraint: (JVET-AL0131 item 4)

For any two different integer values of m and n in the range of 1 to pri_num_resampling_ratios_minus1, inclusive, when value of ( pri_resampling_width_num_minus1[ m ] + 1 ) * ( pri_resampling_width_denom_minus1[ n ] + 1 ) is equal to ( pri_resampling_width_num_minus1[ n ] + 1 ) * ( pri_resampling_width_denom_minus1[ m ] + 1 ), the value of ( pri_resampling_height_num_minus1[ m ] + 1 ) * ( pri_resampling_height_denom_minus1[ n ] + 1 )
and ( pri_resampling_height_num_minus1[ n ] + 1 ) * ( pri_resampling_height_denom_minus1[ m ] + 1 ) shall not be the same.

No action was taken on this.

1. Require uniqueness of region IDs: pri_region_id[ i ] shall not equal to pri_region_id[ j ] when i is not equal to j. (JVET-AL0072 item 1, JVET-AL0129 item 4, JVET-AL0122 item 3)

Adopt item 8.

1. On the value range for pri_region_id[ i ]
2. In VSEI, specify the value range for pri_region_id[ i ] to be in the range of 0 to 2 047, inclusive. (JVET-AL0120 item 10)
3. Specify the value range for pri_region_id[ i ] to be in the range of 0 to 127, inclusive. (JVET-AL0129 item 2)

Adopt a value range limit of 1023.

1. Disallow out-of-cropped-decoded-picture boundary regions.
   1. By adding the following constraints: (JVET-AL0072 item 2)

priRegionTopLeftX[ i ] + priRegionWidth[ i ] shall be less than or equal to MaxPicWidth, and priRegionTopLeftY[ i ] + priRegionHeight[ i ] shall be less than or equal to MaxPicHeight.

• 1. By adding the following constraints: (JVET-AL0122 item 5)

The sum of priRegionTopLeftX[ i ] and priRegionWidth[ i ] shall be less than or equal to PicWidthInLumaSamples[ pri_region_layer_id[ i ] ] and the sum of priRegionTopLeftY[ i ] and priRegionHeight[ i ] shall be less than or equal to PicHeightInLumaSamples[ pri_region_layer_id[ i ] ].

Adopt item 10 option b.

1. Add a note to describe that when target picture is not present, the resampled regions in the current PRI SEI message are unsuitable for human viewing. (JVET-AL0098 item 1)

No action was taken on this.

1. Add an update mechanism for target pic size and region position and size, by adding two flags, pri_target_pic_size_update_flag and pri_region_param_update_flag[ i ]. pri_target_pic_size_update_flag specifies that pri_target_pic_width_minus1 and pri_target_pic_height_minus1 syntax elements are present or not. pri_region_param_update_flag[ i ] specifies that pri_region_top_left_in_units_x[ i ], pri_region_top_left_in_units_y[ i ], pri_region_width_in_units_minus1[ i ], and pri_region_height_in_units_minus1[ i ] syntax elements are present or not. (JVET-AL0098 item 2)

Proposes a partial update mechanism for coding efficiency, which adds dependency on previously signalled SEI messages.

No action was taken on this.

1. Enable signalling of target region position when pri_target_pic_params_present_flag is 0, by adding a flag pri_target_region_pos_present_flag to specify that pri_target_region_top_left_x[ i ] and pri_target_region_top_left_y[ i ] syntax elements are present or not. (JVET-AL0098 item 3)

No action was taken on this.

1. Add derivation of the region parameters for the i-th region when pri_region_is_a_layer_flag[ i ] is equal to 1. (JVET-AL0120 item 2, JVET-AL0303 item 4)

Adopt item 14.

1. Change pri_num_regions_minus1 to pri_num_regions_minus2. (JVET-AL0131 item 7)

This disallows having a single region.

No action was taken on this.

1. Specify the value range for pri_num_regions_minus1 to be in the range of:
2. 0 to 1 023, inclusive. (JVET-AL0120 item 4)
3. 0 to 127, inclusive. (JVET-AL0129 item 1)

Agreed to add a value range of 0 to 255.

1. On coding and value range of pri_region_layer_id[ i ]
   1. In VSEI, specify the value range for pri_region_layer_id[ i ] to be in the range of 0 to 2 047, inclusive, and in VVC interface text, specify the value range to be in the range of 0 to 63, inclusive, in general, and be in the range of 0 to 55, inclusive, in bitstreams conforming to this VVC version. (JVET-AL0120 item 6)
   2. Change the descriptor of pri_region_layer_id[ i ] from ue(v) to u(6). (JVET-AL0129 item 6)

It was suggested that treatment of layer ID should be consistent for messages defined in VSEI v4 and the TuC.

It was agreed to adopt item 17 option a.

1. When the syntax element pri_region_layer_id[ i ] is not present, inferred the value to be equal to the layer identifier of the PU containing the PRI SEI message. (JVET-AL0120 item 7)

Adopt item 18.

1. Signal pri_region_is_a_layer_flag[ i ] regardless of whether pri_multilayer_flag is equal to 0. (JVET-AL0129 item 7)

The proposal would reduce bitrate for a use case with only a single region in a single layer. It was suggested that the same goal could be accomplished with setting the pri_multilayer_flag to 1 even for a single layer.

No action was taken on this.

1. Specify the value range for pri_rsampling_ratio_idx[ i ] to in the range of 0 to pri_num_resampling_ratios_minus1, inclusive. (JVET-AL0120 item 8)

Already addressed in the last version of the spec.

1. Specify a range for the resampling ratio, as follows: (JVET-AL0122 item 1)

( pri_resampling_width_num_minus1[ i ] + 1 ) ÷ ( pri_resampling_width_denom_minus1[ i ] + 1 ) shall be in the range of 1 ÷ 16 to 16, inclusive.

( pri_resampling_height_num_minus1[ i ] + 1 ) ÷ ( pri_resampling_height_denom_minus1[ i ] + 1 ) shall be in the range of 1 ÷ 16 to 16, inclusive.

The proposed range is consistent with VVC RPR resampling ratios. It was suggested to impose the constraint only when target picture parameters are present.

Adopt item 21 with a modification that it is imposed only when pri_target_pic_params_present_flag equal to 1.

1. Disallow overlapping regions in cropped decoded pictures by adding the following constraint: (JVET-AL0122 item 4)

It is a requirement of bitstream conformance that any sample (x, y) in the cropped decoded picture of a specific layer belongs to at most one region.

It was suggested that this is contrary to the original design intention.

No action was taken on this.

1. Clarify the unit and coding length of pri_target_region_top_left_x and pri_target_region_top_left_y by making the following changes: (JVET-AL0122 item 6), as follows:

pri_target_region_top_left_x[ i ] and pri_target_region_top_left_y[ i ], when present, indicate the horizontal and vertical positions, respectively, of the top left sample position in units of the i-th region in the reconstructed target picture. The length, in bits, of the syntax elements are pri_region_size_len_minus1 + 1.

(This replaces “luma samples” with “units”, and adding the second sentence)

Adopt item 23. The editors can also consider if the syntax element should be renamed.

1. Disallow (re)constructed target picture samples out of the target picture.
2. By adding the following constraint: (JVET-AL0072 item 3)

pri_target_region_top_left_x[ i ] shall be in the range of 0 to pri_target_pic_width_minus1, inclusive. pri_target_region_top_left_y[ i ] shall be in the range of 0 to pri_target_pic_height_minus1, inclusive.

1. By adding the following constraint: (JVET-AL0122 item 8)

When pri_target_pic_params_present_flag is equal to 1, the sum of ( pri_target_region_top_left_x[ i ] * priUnitSize ) and priTargetRegionWidth[ i ] shall be less than or equal to ( pri_target_pic_width_minus1 + 1 ), and the sum of ( pri_target_region_top_left_y[ i ] * priUnitSize ) and priTargetRegionHeight[ i ] shall be less than or equal to ( pri_target_pic_height_minus1 + 1 ).

1. By adding the following constraint: (JVET-AL0129 item 12)

pri_target_region_top_left_x[ i ] + priTargetRegionWidth[ i ] and pri_target_region_top_left_y[ i ] + priTargetRegionHeight[ i ] shall be less than or equal to pri_target_pic_width_minus1 + 1 and pri_target_pic_height_minus1 + 1, respectively

Options b and c are similar.

Adopt Item 24 option b.

1. Move pri_region_id_present_flag to include it only when target picture parameters are present. It is asserted that when target picture isn’t reconstructed, there is no need to signal the pri_region_id[ i ] syntax element. (JVET-AL0129 item 5)

No action was taken on this.

1. (No action in JVET-AL_notes_d3) Add the following inference: When not present, the values of pri_target_pic_width_minus1 and pri_target_pic_height_minus1 are inferred to be equal to MaxPicWidth − 1 and MaxPicHeight − 1, respectively. (JVET-AL0070 item 5)
2. (No action in JVET-AL_notes_d3) Un-condition target picture position signalling. (JVET-AL0070 item 6)
3. Add the following constraint to ensure that an entire target region is not covered by another target region: (JVET-AL0131 item 5)

No action was taken on this.

1. Signal chroma format and bit depth for the target picture. (JVET-AL0098 item 4 and JVET-AL0122 item 10)
2. As follows: (JVET-AL0098 item 4)
3. As follows: (JVET-AL0122 item 10)

No action was taken on this.

1. (In JVET-AL_notes_d3: Further discussion requested: If no action is taken on JVET-AL0098 item 4 and JVET-AL0122 item 10, this item should be agreed.) Add the following constraint: ( pri_target_pic_width_minus1 + 1 ) % SubWidthC shall be equal to 0 and ( pri_target_pic_height_minus1 + 1 ) % SubHeightC shall be equal to 0. (JVET-AL0070 item 8)

Adopt item 30.

1. On signalling of region parameters in a chroma-aligned manner
2. Add one of the following two constraints:
   • 1. priRegionTopLeftX[ i ] % SubWidthC shall be equal to 0 and priRegionTopLeftY[ i ] % SubHeightC shall be equal to 0. (JVET-AL0070 item 9 solution 1 part 1)
     2. priRegionTopLeftX[ i ] % SubWidthC[ pri_region_layer_id[ i ] ] shall be equal to 0 and priRegionTopLeftY[ i ] % SubHeightC[ pri_region_layer_id[ i ] ] shall be equal to 0, where SubWidthC[ pri_region_layer_id[ i ] ] and SubHeightC[ pri_region_layer_id[ i ] ] are set equal to SubWidthC and SubHeightC, respectively, when ChromaFormatIdc is equal to ChromaFormatIdc[ pri_region_layer_id[ i ] ]. (JVET-AL0098 item 4)
     3. priRegionTopLeftX[ i ] % SubWidthC[ lId ] shall be equal to 0 and priRegionTopLeftY[ i ] % SubHeightC[ lId ] shall be equal to 0. (JVET-AL0129 item 11)

All 3 items are essentially the same, but editorially different.

Adopt item 31 a option i.

1. Add one of the following two constraints:
   • 1. pri_target_region_top_left_x[ i ] % SubWidthC shall be equal to 0 and pri_target_region_top_left_x[ i ] % SubHeightC shall be equal to 0. (JVET-AL0070 item 9 solution 1 part 2)
     2. priTargetRegionTopLeftX[ i ] % TargetSubWidthC shall be equal to 0 and priTargetRegionTopLeft[ i ] % TargetSubHeightC shall be equal to 0, where TargetSubWidthC and TargetSubHeightC are set equal to SubWidthC and SubHeightC, respectively, when ChromaFormatIdc is equal to pri_target_pic_chroma_format_idc, priTargetRegionTopLeftX[ i ] is equal to ( pri_target_region_top_left_x[ i ] + 1 ) * priUnitSize, and priTargetRegionTopLeftY[ i ] is equal to ( pri_target_region_top_left_y[ i ] + 1) * priUnitSize. (JVET-AL0098 item 4)
     3. ( pri_target_region_top_left_x[ i ] * priUnitSize % SubWidthC ) and ( pri_target_region_top_left_y[ i ] * priUnitSize % SubHeightC ) shall be eqaul to 0. (JVET-AL0122 item 7)

All 3 items are essentially the same, but editorially different.

Adopt item 31 b option iii.

1. If no to a and b above, do either of the following:
   • 1. Related variables for region parameters are translated by using SubWidthC and SubHeightC. (JVET-AL0070 item 9 solution 2)
     2. Use the mechanism as the annotated regions SEI message and the object mask information SEI message for signalling of region parameters. (JVET-AL0070 item 9 solution 3)
2. Update the following constraints as highlighted: It is a requirement of bitstream conformance that priRegionWidth[ i ] % SubWidthC[ pri_region_layer_id[ i ] ] shall be equal to 0 and priRegionHeight[ i ] % SubHeightC[ pri_region_layer_id[ i ] ] shall be equal to 0. (JVET-AL0098 item 4)

No action on c and d.

1. Specify the derivation of priTargetRegionWidth and priTargetRegionHeight when pri_use_max_dimensions_flag is true: (Adopt JVET-AL0303 item 6)
2. Fix the bug related to target picture (re)construction. (JVET-AL0070 item 7, JVET-AL0122 item 11, JVET-AL0129 item 3)
3. Change the following (JVET-AL0070 item 7)
4. Make the following changes (JVET-AL0122 item 11):
5. Make the following changes: (JVET-AL0129 item 3)

Adopt item 33 option b.

1. Add signalling of human viewing and machine analysis suitability, as follows: (JVET-AL0131 item 6)

No action was taken on this.

1. Editorial changes that could be delegated to the editors:
2. (Agreed in JVET-AL_notes_d3) JVET-AL0070 item 2
3. (Agreed in JVET-AL_notes_d3) JVET-AL0070 item 3 (JVET-AL0132 item 2 also proposes).
4. Typo fixes in JVET-AL0070 item 9: MaxWidth ->MaxPicWidth, MaxHeight -> MaxPicHeight
5. JVET-AL0120 item 9, JVET-AL0303 item 1
6. JVET-AL0132 item 3

Delegate item 35 to the editors.

Decision: Adopt as noted as agreed above.

Image format metadata SEI message (3+1)

Contributions in this area were discussed during 1300–1410 on Tuesday 1 April 2025 (chaired by S. Deshpande).

JVET-AL0324 (incl. aspects of JVET-AL0072, JVET-AL0098, JVET-AL0120, JVET-AL0122, JVET-AL0129, JVET-AL0303) AHG9: A summary of proposals on the PRI SEI message