Back to Search Document details
39th Meeting: Daejeon, KR, March 2025 2025-03-27 12:45
AHG9: Further fixes and cleanup on GFV and GFVE SEI messages
Abstract
This contribution fixes and cleanup the current semantics of generative face video and generative face video enhancement SEI messages in VSEI spec. (JVET-AK2006) and also proposed to modify a SEI message related constraint in VVC spec.
JVET-AL0155 AHG9: Further fixes and cleanup on GFV and GFVE SEI messages [J. Chen, B. Chen, Y. Ye (Alibaba)]

Aspect 1: remove the inference for the syntax elements gfv_num_matrices_equal_to_num_kps_flag[ i ], gfv_num_matrices_info[ i ], gfv_matrix_width_minus1[ i ], gfv_matrix_height_minus1[ i ] gfv_num_matrices_minus1[ i ] and gfv_matrix_for_3D_space_flag[ i ] in case of prediction being enabled.

Aspect 2: modify the inference for the syntax element gfv_matrix_for_3D_space_flag[ i ] to align the semantics and simplify the inference for the syntax elements gfv_matrix_width_minus1[ i ] and gfv_matrix_height_minus1[ i ].

Aspect 3: modify the inference condition for the syntax elements gfv_matrix_element_precision_factor_minus1 and gfv_num_matrix_types_minus1 for text cleanup.

Aspect 4: remove redundant process in the derivation of variable numMatrices[ i ].

Aspect 5: remove unnecessary constraint on gfv_num_matrix_types_minus1

Aspect 6: remove unnecessary constraint on numMatrices[ i ], matrixWidth[ i ] and matrixHeight[ i ].

Aspect 7: add the inference for the syntax elements gfve_matrix_element_precision_factor_minus1, gfve_num_matrices_minus1, gfve_matrix_height_minus1[ i ] and gfve_matrix_width_minus1[ i ] which is currently missing from the text.

Aspect 8: change a constraint in VVC spec. to allow signalling multiple GFV and GFVE SEI messages with different content within one picture unit.

Aspect 1, Aspect 2, Aspect 4 agreed. (Editorial)

For aspect 3, the change to semantics of gfv_matrix_element_precision_factor_minus1 is agreed, with addition of removing “when gfv_matrix_present_flag is equal to 1”. For the semantics of gfv_num_matrix_types_minus1, can just say when “gfv_matrix_pred_flag is equal to 1". Aspect 3 agreed with noted changes and editorial cleanup delegated to the editors.

Further discussion requested onaspect 5, aspect 6.

Aspect 7, Aspect 8 agreed.

Further discussed on 1 April 2025.

Regarding aspect 5 and 6, the meaning of the word “each” in the constraint that is being removed was unclear.

Aspects 5 and 6 agreed.

Decisions
Aspects 5 and 6 agreed.
Citation