International organisation for standardisation organisation internationale de normalisation

6TE4: Variable length coding

6.1.1.1.1.1.1.1.1JCTVC-C261 TE4: Summary of TE4 on variable length coding [X. Wang (Qualcomm)]

This document summarized the activities in the Tool Experiment TE4 on Variable Length Coding (VLC). A group of five companies and universities had registered for participation in TE4. One TE report, JCTVC-C262, had been submitted on VLC coding for coded block flags.

6.1.1.1.1.1.1.1.2JCTVC-C262 TE4: Report on VLC for coded block flag [X. Wang, M. Karczewicz, W.-J. Chien (Qualcomm)]

Document JCTVC-C262 reported TE results based on a coding scheme presented in a previous proposal JCTVC-B098. In JCTVC-B098, coding schemes for coded block flag both at CU level and at block level were proposed. Due to the fact that the proposed coded block flag coding at CU level is very close to the one introduced in "phase 2" VLC integration, JCTVC-C262 further reported results of coded block flag coding only at block level.

According to the proposal, if a coded block flag for a certain video component at the coding unit level is not zero and the transform block size is smaller than the coding unit size, the coded block flags from the four quarter-sized blocks are grouped and coded together. Due to coded block flag at the coding unit level, the coded block flags for the four quarter-sized blocks cannot all be zero. As a result there are 15 different combinations in total for these flags. A new VLC table was designed and tested. It was reported that on average around 0.2% coding gain can be obtained using the three low complexity coding configurations.

It was noted that the proposal has an interaction with the use of a residual quadtree. The proposal does not apply if residual quadtree encoding with more than two levels is used.

The proposal only applies to the LCEC entropy coding scheme.

The coding gain is rather minor, but the proposal is a small change.

It was suggested that if the proposal is adopted, it should be possible to switch it off in the software, to ease future testing.

Decision: It was agreed to adopt the proposal into the TMuC / TM (provided it does not conflict with other adoptions).

See also the discussion of JCTVC-C277 and JCTVC-C319.

6.1.1.1.1.1.1.1.3JCTVC-C288 TE4: Verification results of JCTVC-C262 [T. Yamakage (Toshiba)] (missing prior, available first day)

This contribution shows verification results of JCTVC-C262 (Report on VLC for coded block flag) for TE4. The contributor started with source code and compiled it themselves to produce executables, but did not study the software algorithmically. Results were reported to be slightly different but not significantly, relative to JCTVC-C262. No mismatch between the encoder and decoder was observed. Encoding and decoding time comparison with the TMuC 0.7.4 anchor (including macro change) showed approximately no impact.

7TE5: Simplified TMuC intra prediction

7.1.1.1.1.1.1.1.1JCTVC-C046 TE5: Summary report on simplification of unified intra prediction [T. K. Tan (NTT DoCoMo), M. Budagavi (TI), J. Lainema (Nokia)]

This is the summary report for the TE5 on Simplification of Unified Intra Prediction. The description of the experiment can be found in JCTVC-B305r1. This tool experiment was successfully conducted and verified the simplification proposed in JCTVC-B093 and JCTVC-B118 within the context of the unified intra prediction method proposed in JCTVC-B100. Simulation results reportedly show that both methods do not result in a significant loss in coding efficiency compared to the TMuC 0.7.

In the angular intra prediction method, two arrays of reference samples are used, corresponding to a row of samples lying above the current block to be predicted, and a column of samples lying to the left of the same block. Given a dominant prediction direction (horizontal or vertical), one of the reference arrays is defined to be the "main" array and the other array the "side" array. In the prediction process, only samples from the main array are used for prediction when the angle value is positive. When the angle value is negative a per-sample test is performed to determine whether samples from the main or the side array should be used for prediction. Additionally when the side array is used, the index into the array is obtained by a mapping operation that requires either a division operation or a table look-up into a sizable table.

JCTVC-B093 simplifies the prediction process when the angle value is negative. The main array is extended by projecting samples from the side array onto it according to the prediction direction. This projection entails copying a subset of the samples from the side array into the main array. Note that in the copy process, the inverse angle may be replaced with a lookup table. During the prediction process, only the extended main array is used and the same simple linear interpolation formula is used to predict all samples in the block.

JCTVC-B118 replaces divisions in angular intra prediction with a lookup table and additions.

All six coding conditions cases were tested.

Decision: After discussion of the relevant contributions, it was agreed to adopt these two simplifications into the TMuC and TM.

7.1.1.1.1.1.1.1.2JCTVC-C042 TE5: Results for simplification of unified intra prediction [T. K. Tan, Frank Bossen (NTT DoCoMo)]

This contribution reported simulation results for Tool Experiment 5 as described in JCTVC-B305r1. Two tools were compared. The first tool, based on JCTVC-B093, simplifies the angular intra prediction by eliminating the sample-based checking and division steps. The second tool, based on JCTVC-B118, replaces the division in angular intra prediction with a lookup table and additions.

The results reportedly confirmed that both tools do not change the objective performance of the angular prediction process in any significant way. However, the simplified angular prediction (JCTVC-B093) provides the additional advantage of removing bottlenecks in the reference TMuC algorithm to enable a lower-complexity method for intra prediction for which parallel processing can be achieved for all prediction angles on typical SIMD architectures.

7.1.1.1.1.1.1.1.3JCTVC-C061 TE5: TI evaluation of unified intra prediction simplifications [M. Budagavi (TI)]

In HEVC TMuC-0.7 Intra prediction, pixels in current block are predicted from their neighboring left (side reference) and top pixels (main reference) using angular intra prediction. Prediction of pixels from side reference involves division of angle which is computationally complex to implement. Two proposals were made at the last JCT-VC meeting to simplify intra prediction from side reference – JCTVC-B118 and JCTVC-B093. This contribution presents results of evaluation of these two proposals for simplification of intra prediction. The simplifications are achieved using the following two tools: division elimination (JCTVC-B118, JCTVC-B093) and fractional pixel position calculation optimization (JCTVC-B093). Based on the results presented in this contribution, the contributor recommended that these two tools be adopted into TMuC and Test Model (TM).

7.1.1.1.1.1.1.1.4JCTVC-C100 TE5: Huawei & HiSilicon report on verification test results [L. Liu (HiSilicon), J. Zhou, H. Yu (Huawei)]

From the results it was confirmed that the proposed methods do not result in any significant difference in coding efficiency compared to the TMuC 0.7.

Both simplified angular prediction (JCTVC-B093) and JCTVC-B118 successfully eliminate the division using a lookup table.

Simplified angular prediction also has the additional benefit of being more suited for typical SIMD architectures due to the single reference array and a single equation for the creation of all predicted samples in the block.

It was recommended that the division elimination and sample creation method of simplified angular prediction be added to the TMuC and Test Model (TM).

7.1.1.1.1.1.1.1.5JCTVC-C186 TE5: Results for simplified intra prediction tests by Nokia [J. Lainema, K. Ugur (Nokia)]

This document reported Nokia results for the TE5 tests and verifies the correctness of the results submitted by other TE participants.

This contribution was based on a separate independent implementation of the proposal.