International organisation for standardisation organisation internationale de normalisation

8.4TE6.a.3 Overlapped block intra prediction

8.4.1.1.1.1.1.1.1JCTVC-C193 Overlapped block intra prediction [K. Zhang, X. Guo, J. An, M. Guo, Y.-W. Huang, S. Lei (MediaTek)]

This contribution presented the results of overlapped block intra prediction (OBIP) in Tool Experiment TE6.a. In the proposed OBIP method, a position-dependent weighted sum of several possible predictors is treated as the final prediction. The weighting values are obtained by offline training and are fixed during the encoding/decoding procedure. The OBIP method had been implemented into TMuC software. Experimental results reportedly showed that OBIP can attain an average bit-rate reduction of 1% and 1.7% for all-intra high efficiency and low complexity, respectively.

The technique uses a position-dependent weighted combination of three predictors. The weighting combination is selected from one of nine weighting matrices. Only 4x4 and 8x8 blocks use this (as implemented).

Edge-based prediction and planar prediction were turned off (which reportedly has essentially negligible PSNR impact) for reasons of tool incompatibility.

The non-overlapped modes were retained in the scheme as well as the additional OBIP modes.

Further study was proposed, and encouraged.

8.4.1.1.1.1.1.1.2JCTVC-C272 TE6: Cross-verification of MediaTek's overlapped block intra prediction by Microsoft [X. Peng, J. Xu (Microsoft)] (missing prior – later provided)

This contribution reported cross-verification test results for OBIP as proposed in JCTVC-C193. The results reportedly matched those reported in JCTVC-C193.

The software was studied as well as compiled and run.

8.5TE6.a.4 Line-based intra prediction

8.5.1.1.1.1.1.1.1JCTVC-C270 TE6: Report of line-based coding [X. Peng, J. Xu, F. Wu (Microsoft)]

This document reported experimental results of the line-based coding scheme, which was first presented in JCTVC-A118. The line-based coding takes a line as the reconstruction unit and utilizes content-adaptive prediction filters for accurate predictions. When integrated into the TMuC0.7 software, it reportedly showed 2.1% and 2.4% bit rate saving on average, under all intra high efficiency and low complexity conditions, respectively. Up to 5.1% bit rate saving was reportedly achieved on sequences with rich edges.

Line lengths of 4, 8, and 16 were tested. Syntax is not used to indicate the prediction parameters – rather, they are derived based on the reconstructed area.

The reported encoding and decoding times were approximately doubled or tripled.

It was remarked that the complexity impact of the proposed scheme is substantial.

Further study was encouraged.

Also see remarks above regarding JCTVC-C101.

8.5.1.1.1.1.1.1.2JCTVC-C045 TE6.a: Cross verification for line based intra prediction [T. K. Tan, Junya Takiue (NTT DoCoMo)]

This contribution was also described above. It contained cross verification reporting for both JCTVC-C101 (JCTVC-B040) and JCTVC-C270.

8.5.1.1.1.1.1.1.3JCTVC-C139 TE6.a: Huawei & HiSilicon report on cross verification of MSRA's LIC [C. Lai, Y. Lin (HiSilicon), J. Zhou (Huawei)]

This contribution provided a summary of the cross-checks performed by Huawei Technologies and Hisilicon Technologies on MSRA’s LIC (Line Intra Coding) proposal. The simulation results reportedly exactly matched the results in terms of bit rate and PSNR as provided by Microsoft.

8.6TE6.b.1 Differential coding of intra modes (DCIM)

8.6.1.1.1.1.1.1.1JCTVC-C169 TE6.b: Experiment results of DCIM [E. Maani, L. Dong, W. Liu (Sony)]

This document presented experimental results of Differential Coding of Intra Modes (DCIM) under the test conditions defined for Tool Experiment 6.

DCIM uses neighborhood edge estimation to predict Intra prediction direction and differentially encodes the modes with respect to the predicted direction. This enables a higher precision in the Intra prediction directions without substantially increasing the mode signaling overhead. An additional bit is transmitted per block (i.e., PU) to signal to the decoder weather DCIM is used.

MDDT was disabled due to tool implementation compatibility issues, although ROT was used.

Approximately 1.9% overall gain was reported.

For the LC configuration, the work was affected by a bug (bug fix 79) in the underlying TMuC software. The cross verifications with the bug fix had not yet been completed at the time of presentation.

Encoding time is approximately doubled (possibly somewhat slowed by increased use of ROT due to disabling of MDDT); decoding time is reportedly decreased by about 40% (due to disabling of MDDT).

The contributor recommended further study, which was encouraged.

8.6.1.1.1.1.1.1.2JCTVC-C175 TE6.b: Cross verification of differential coding of intra modes (DCIM) [T. Yamamoto, Y. Yasugi (SHARP)] (missing prior, available first day)

This contribution provided cross-check results of the method called Differential Coding of Intra Modes (DCIM) proposed by Sony in the context of TE6 (TE6.b: Edge based intra prediction). The experiments defined in TE6 were carried out using the software provided by Sony. The experimental results reportedly showed that the proposed method provides 1.9% gain in High Efficiency / Intra Only operation. It was reportedly confirmed that these results matched the results provided by the proponent.

Testing for the LC configuration context had encountered difficulties as described above (bug fix 79).

It was suggested that perhaps, in general, we should not use the "LC" configurations to test techniques that require a substantial complexity increase.

8.6.1.1.1.1.1.1.3JCTVC-C217 TE06.b: Cross verification of differential coding of intra modes (DCIM) [V. Drugeon (Panasonic)]

This contribution was a cross-check for differential coding of intra modes (DCIM). Partial results were provided. The results reportedly closely matched those reported by Sony. The software was studied and determined to match the algorithm description as proposed in July.

8.6.1.1.1.1.1.1.4JCTVC-C176 Analysis and improvement of differential coding of intra modes [T. Yamamoto, Y. Yasugi (Sharp)]

This contribution related to the DCIM technology tested in TE6.

Edge based intra prediction is a method to utilize the gradient of the neighboring samples to decide the direction to predict the target partition. Since one of its variants has been incorporated in the TMuC and the other variant (DCIM: Differential Coding of Intra Modes) was being evaluated in a TE, the Edge based intra prediction was one of the actively researched areas. In this contribution, some analysis of DCIM was presented for providing a better understanding of the technology. Also, some modification and (limited) experimental results were reported – with reported benefits of approximately 0.2% in HE and 0.7% in LC configurations.

It was asked whether the parsing of the bitstream is dependent on the values of the reconstructed samples. The contributor said that it does not.