Joint Collaborative Team on Video Coding (jct-vc)
Reference picture set (8 – done)
Yüklə 1,12 Mb.
|
- Bu səhifədəki naviqasiya:
- JCTVC-J0185 AHG13: On short_term_ref_pic_set [S. Lu, K. Sato (Sony)]
- JCTVC-J0517 Mental cross-check of JCTVC-J0185 (On short_term_ref_pic_set) [Chong Soon Lim (Panasonic)] [late]
- JCTVC-J0545 Mental cross-check of JCTVC-J0234: Inter-RPS complexity reduction [TK Tan (NTT Docomo)] [late]
- JCTVC-J0402 Crosscheck of Signalling of long-term reference pictures in the slice header in JCTVC-J0115 [Hendry, B. Jeon (LG)] [late] [miss]
- JCTVC-J0116 AHG13: Signalling of long-term reference pictures in the SPS [A. K. Ramasubramonian, Y.-K. Wang, Y. Chen (Qualcomm), C. S. Lim (Panasonic), S. Deshpande (Sharp) , Hendry (LG)]
- JCTVC-J0480 Crosscheck of long term reference pictures in SPS (JCTVC-J0116) [TK Tan ()] [late]
- JCTVC-J0538 Mental Cross-check for JCTVC-J0118 [S. Deshpande (Sharp)] [late] JCTVC-J0164 AHG 13: Simplified signalling of MSB cycle [Hendry, B. Jeon (LG)]
- JCTVC-J0539 Mental Cross-check for JCTVC-J0164 [S. Deshpande (Sharp)] [late]
5.12.4Reference picture set (8 – done)JCTVC-J0513 Common conditions for reference picture marking and list construction proposals [R. Sjöberg, Y.-K. Wang, M. Hannuksela, T. K. Tan, Y. Ye] [late] TBA. This was the result of interim AHG activity. This was approved as the appropriate method of testing typical characteristics for proposals in this area (for future proposals as well as prior ones). JCTVC-J0185 AHG13: On short_term_ref_pic_set [S. Lu, K. Sato (Sony)] Short_term_ref_pic_set was asserted to contain the following redundancies:
In this contribution it is proposed to modify syntax related to short_term_ref_pic_set to remove above redundancies. This was closely related to J0234. See the notes in the section on that document. Cross-check in J0517.
This contribution proposes two changes to how inter-RPS parameters are encoded, and one change that removes an alleged underspecification. The first proposed change is to not transmit the variable delta_idx_minus1 when encoding inter-RPS parameters in the SPS, and instead infer that it has the value 0. The variable is still transmitted when encoding inter-RPS parameters from the slice header. The second proposal is to signal a flag called used_by_curr_present_in_inter_rps_flag in the SPS. If this flag is false, all inter-RPS coded reference pictures are assumed to have used_by_curr_pic_flag[ j ] = 1 and this means that the flag does not need to be transmitted. Both proposals have been tested on the entire AHG13 set of configuration files as specified in JCTVC-I0608. The proposal claims to lower the average complexity slightly and reduce the number of bits spent on sending the RPS data in the SPS with −11.8%. Separately, the first and second proposals are reported to reduce the number of bits by −7.0% and −4.8% respectively. The third proposal is to add semantics for what happens if the RPS row to predict from does not exist; then the empty set will be used for prediction. Cross-check provided by Docomo in J0545. Several participants expressed a preference for elements #1 and #3 of this proposal relative to the current draft and relative to J0185. The element #2 of the contribution was less well supported. Tentative plan is to adopt #1 and #3 (and not #2) of this proposal. This was discussed further later. After offline consideration, it was suggested in regard to #3 that, rather than having the concept of prediction from an empty set, there should be a constraint imposed to prohibit a reference to a non-existing set. It was indicated that this would be described in a revision of J0185.
It was reported that there is a problem with the signalling of long-term reference pictures (LTRPs). An error-resilience problem is reported for the current signalling and derivation of long-term reference pictures (LTRPs). It is stated that the root of problem was that the reference picture set (RPS) derivation depends on the status of the decoded picture buffer (DPB), which was the main reason for replacing the sliding-window and memory management control operation (MMCO) based reference picture buffer management mechanism in AVC with the RPS based mechanism. This document proposes to send the delta POC values, either between the current picture and the LTRP or between the LTRP and the previous random access point (RAP) picture in decoding order to solve the reported error resiliency issue, and to satisfy the RPS design principle that the RPS derivation is self-contained (i.e., not depending on the DPB status). The simulation results are in the attachment of the proposal document. Two cross-checks planned (J0402 and J0530), but not yet submitted when reviewed. J0402 was not provided, and was therefore considered withdrawn. The current syntax can (only) have some case of non-robust behaviour if the encoder chooses to put it in that mode (not sending MSB cycle difference) and re-uses the same POC LSBs for multiple LTRPs. The proposal would require always sending the MSB cycle difference rather than allowing the encoder to make this choice, and offers a way to code that difference more efficiently if the LTRP is close in position to the preceding RAP. These two aspects can, in principle, be considered as two separate suggestions. There was no consensus that the proposed changes were appropriate to do. No action taken.
This document proposes to enable the inclusion of candidate long-term reference pictures, as part of the reference picture set signalling in the sequence parameter set. Two different options are provided. The document reports that, for test condition 2.6 in JCTVC-H0725, the proposed method uses 36% and 42% fewer bits, respectively for the two proposed options, to signal the syntax elements related to long-term reference pictures in the sequence parameter set and the slice header, when compared to the signalling in HEVC text specification draft 7. The first presented option was a design considered at the preceding meeting in I0340, for which further study had been suggested. The second was a design that had also been in a version of I0340. (A four-company proposal – also supported by additional non-proponents.) Decision: Adopt "option 2" with u(v) coding of lt_idx_sps. JCTVC-J0480 Crosscheck of long term reference pictures in SPS (JCTVC-J0116) [TK Tan (??)] [late] JCTVC-J0118 AHG13: On signalling of MSB cycle for long-term reference pictures [A. K. Ramasubramonian, R. L. Joshi, Y.-K. Wang (Qualcomm)] This document proposes changes to the semantics of the syntax elements poc_lsb_lt[ i ], delta_poc_msb_present_flag[ i ], and delta_poc_msb_cycle_lt[ i ] to improve the efficiency of signalling the MSB cycle for long-term reference pictures in the slice header. Changes to the RPS derivation process are also proposed, including swapping of the order of STRP and LTRP subset derivations, such that the STRP subset is derived first. If the proposal in JCTVC-J0115 is adopted, then this proposal is claimed to become irrelevant and should be ignored. No experiment results were provided. To use the suggested scheme, the specification of the decoder process for building the reference picture set would need to be change. There were some problems in the text that was provided with the contribution for these changes. A revision of the contribution was provided to address that. The proposed text may also include some editorial improvements of the expression of the existing scheme, and it was suggested for the editors to consider any such improvements. This was agreed. Decision (Ed.): Editor action item. No action taken.
In principle, there are situations in which the decoder can recognize that the encoder will send the POC MSB cycle difference. The basic idea of the proposal is to eliminate the overhead of indicating whether the cycle difference will be sent in such a situation. However, there were problems in some of the specific changes that were proposed. No action taken. JCTVC-J0539 Mental Cross-check for JCTVC-J0164 [S. Deshpande (Sharp)] [late] Yüklə 1,12 Mb. Dostları ilə paylaş:
|