International organisation for standardisation organisation internationale de normalisation

9.4Toshiba proposed simplification of MDDT

9.4.1.1.1.1.1.1.1JCTVC-C080 TE7: One-dimensional directional unified transform [A. Tanizawa, J. Yamaguchi, T. Shiodera, T. Chujoh, T. Yamakage (Toshiba)]

This contribution presented experimental results of 1 Dimensional Directional Unified Transform (1DDUT) for Tool Experiment 7 (MDDT simplification in HEVC). 1DDUT is a method of reducing the number of transform matrices of MDDT by using a set of pre-defined 1 D transform matrices depending on the intra prediction direction. 1DDUT was included in the CfP submission of JCTVC-A117 and the improved method was proposed in the contribution of JCTVC-B042 at JCT-VC Geneva meeting.

In this document, the experimental results in TMuC software version 0.7.1 on the high efficiency conditions based on the common test conditions for both I slice only coding structure and the random access coding structure defined by TE7 were reported. For I slice only coding structure, the BD-Bit rate gain compared with the anchor (MDDT on) was reported to be 0.1% on average, and for the random access coding structure, the BD-Bit rate gain was reported as 0.0% on average.

9.4.1.1.1.1.1.1.2JCTVC-C035 TE7: Cross-verification of the 1D directional unified transform in TMuC [R. Cohen, A. Vetro, H. Sun (Mitsubishi)]

This document reported the outcome of Mitsubishi Electric Research Laboratory’s (MERL) cross-verification of Toshiba’s 1D Directional Unified Transform (1DDUT) in TMuC. This activity is part of Tool Experiment 7 on MDDT simplification. A brief analysis of the architecture and software was followed by a summary of experimental results. Differences (negligible, averaging 0.0%) in bit rates and PSNR values can be accounted for by the difference in simulation platforms. The conclusion is that the MERL cross-verification results match those produced by Toshiba.

The software was studied and indicated to be straightforward and to reflect what is described in JCTVC-C080, although a bit different (simpler) than what was proposed in Geneva.

9.4.1.1.1.1.1.1.3JCTVC-C103 TE7: Report of cross-checking results for Toshiba proposal [H. Yang, J. Zhou, H. Yu (Huawei)]

This document reported cross checking results for Toshiba’s proposal in TE 7 activities. The decoded sequences matched the reconstructed sequences. The RD data provided by Toshiba were reportedly the same as the results obtained from the crosschecking (with only very minor differences likely due to compiler and platform issues).

9.5Beijing Univ. Tech. proposed simplification of MDDT

9.5.1.1.1.1.1.1.1JCTVC-C305 TE7: Results for Simplification of MDDT Transform [W. Ding, Y. Shi, B. Yin (Beijing Univ. Tech.)] (late)

MDDT has been implemented as a direct matrix multiplication, which results in higher computation cost than that of an ordinary integer transform. This contribution proposed two symmetric transform matrices to simplify the MDDT matrices. The proposed symmetric transform matrices can be implemented using a lifting structure. Experimental results reportedly demonstrate that this proposal reduced the run-time complexity of MDDT with approximately 0.5% coding performance loss.

The operation count was asserted to be lower in this contribution than in the others, by enforcing symmetric transform matrices, which may be why the loss is a bit larger.

The contribution was submitted late, and no cross check was available.

9.6Peking Univ. proposed simplification of MDDT

9.6.1.1.1.1.1.1.1JCTVC-C089 TE7: Results for mode-dependent residual reordering for intra prediction residual [X. Zhao, L. Zhang, S. Ma, W. Gao (Peking Univ.)]

In this contribution, the Peking Univ. results in response to the Tool Experiment TE7 on MDDT simplification were presented, with the tool description and simulation results. To simplify the mode-dependent directional transform (MDDT), a mode-dependent residual reordering (MDRR) method was proposed. In the proposed MDRR, between the prediction and transform stages, a mode-dependent reordering is implemented on the residual samples for each mode in the spatial domain. After the reordering, a single transform matrix is applied. In all-intra HE configuration, a 0.3% average loss was reported.

9.6.1.1.1.1.1.1.2JCTVC-C027 TE7: Verification results of Peking Univ. proposal (mode-dependent residual reordering for intra prediction residual) [J. Kim, B. Jeon (LG Electronics)]

This document reported verification results of JCTVC-C089 on mode-dependent residual reordering for intra prediction residual, submitted by Peking University. The proponents provide their source code for cross-checking.

For HE intra, the simulation succeeded on all platforms of Window7 32bit and WindowXP 64bit and reportedly produced the same results as what the proponent provided. For HE random access, the simulation succeeded and reportedly produced the same results as what the proponent provided when the same platform was used as what the proponent used.

An accompanying spreadsheet shows different results for V-PSNR for BasketballDrive_1920x1080_QP22 but it was reported to seem to be caused by minor rounding issues. The results of HE random access coding on Windows 7 32bit and on WindowXP 64bit have differences for some sequences such as BQsquare_416x240. An encoder crash occurred for some sequences such as Racehorse_832x480_QP37 on Windows XP 64bit.

The software was not studied – just compiled and run.

9.6.1.1.1.1.1.1.3JCTVC-C081 TE7: Cross-verification of mode dependent residual reordering (Peking Univ.) [A. Tanizawa, J. Yamaguchi (Toshiba)]

This contribution contained cross-checking of software containing the tool "Mode-dependent residual reordering (MDRR)" of JCTVC-B102 proposed by Peking University. The verification task was reported to have been completed successfully and the results reportedly matched those reported in JCTVC-C089. The encoded images matched those of the decoded ones, and the PSNR and bit rate values were reportedly identical to the results provided by Peking University except for one case. In the case of RaceHorses (QP=37), an encoder crash happened. However, the contributor suggested that this might be caused by a bug in the memory access in TMuC version 0.7.

The software was not studied – just compiled and run.

9.6.1.1.1.1.1.1.4JCTVC-C190 TE7: Cross-check result of Peking Univ.'s proposal (JCTVC-C089) [J. Kim, S.-C. Lim, H. Y. Kim, H. Lee, J. S. Choi (ETRI)]

This document presented verification results of JCTVC-C089 proposed by Peking University. Some encoder crashes were experienced. When the encoder did not crash, matching results were obtained.

The cross-checker verified that no unrelated areas of the software that had been modified, but did not closely study exactly what the proposal part of the software was doing algorithmically.