Organisation internationale de normalisation

SCC performance, design aspects and test conditions (2)

(Consideration of this topic was chaired by Rajan Joshi on Thursday 06-25, 18:30-19:00.)

1.1.1.1.1.1.1.1.22JCTVC-U0051 Comparison of Compression Performance of HEVC Test Model 16.4 and HEVC Screen Content Coding Extensions Test Model 4 with AVC High 4:4:4 Predictive profile [B. Li, J. Xu, G. J. Sullivan (Microsoft)]

This contribution reports a study of the relative objective (i.e. PSNR-based) compression performance of HEVC RExt Main 4:4:4 Profile, HEVC Screen Content Coding (SCC) Test Model 4 (SCM 4) and AVC High 4:4:4 Predictive Profile. It builds upon the prior work reported in JCTVC-G399, JCTVC-H0360, JCTVC-I0409, JCTVC-J0236, JCTVC-K0279, JCTVC-L0322, JCTVC-M0329, JCTVC-O0184, JCTVC-P0213, JCTVC-R0101, JCTVC-S0084, and JCTVC-T0042 – updating the results by using the latest available reference software (JM-18.6, HM-16.4, HM-16.4+SCM-4.0), profile and test model designs, and SCC common test conditions (CTC) test sequences. The overall results reportedly indicate that for screen content CTC sequences, the HEVC SCC Test Model 4 improves quite substantially over JM-18.6. For example, for RGB text and graphics with motion (TGM) 1080p & 720 sequences, it is reported that the BD-rate savings of the HEVC SCC Test Model 4 over JM-18.6 are 86%, 80%, and 78% for lossy AI, RA and LB configurations, respectively.

The reported BD-rate gains for 4:4:4 sequences are similar to the previously reported gains. Reported gains for the 4:2:0 sequences are larger. It is asserted that this is due to the addition of new 4:2:0 test sequences and enabling of palette mode for 4:2:0. The decoding times are reportedly not very accurate but were asserted to give an overall indication. The BD-rates were obtained using rate matching. Lossless gains are smaller in general. It was commented that there may be an error in lossless results (vs JM) for 4:2:0 sequences. HM16.4 used for the comparisons did not contain the encoder changes specific to screen content material that are present in HM 16.4 + SCM 4.0. One participant asked whether 10-bit sequences coded with JM-18.6 in 8-bit mode were handled similarly to HM 16.4 + SCM 4.0. The presenter was not sure about this.

1.1.1.1.1.1.1.1.23JCTVC-U0058 Compression Performance of HEVC Screen Content Coding Extensions Test Model 4.x with slices [C. Gisquet, G. Laroche, P. Onno (Canon)] [late]

This contribution is a study of the relative objective (i.e. PSNR-based) compression performance of HEVC SCC Test Model 4.x (SCM 4.x) using a configuration of 1500 bytes per slice. At the time of writing, SCM4.1 is still being under development, but the present contribution is asserted to provide relevant results.

The modified anchor used a version of SCM4.x (revision 4482) with 1500 bytes per slice, where reference samples for intra block copying (IBC) are restricted to be within the current slice. A bug related to reference picture list management is also fixed in the modified anchor and the palette predictor initializer is reported set to 1.

The BD-rate performance of the modified anchor is reported for configurations where with one tool disabled at a time (ACT, IBC, palette). It is reported that for RGB text and graphics with motion (TGM) 1080p & 720 sequences for the AI configuration, the BD-rate losses due to disabling ACT, IBC and palette are 9.6%, 18.7%, and 45.2%, respectively. For YUV text and graphics with motion (TGM) 1080p & 720 sequences, the corresponding BD-rate losses were reported to be 0.2%, 23.0% and 44.1%, respectively. It was asserted that in contrast, without the use of slices, the BD-rate losses due to disabling IBC are much higher than disabling palette.

It was commented by the presenter that compared to palette and IBC, ACT adds more run-time complexity even in case of random access and low delay configuartions. A participant commented that this was partly due to the fact that efficient hash-based searches are used for inter and IBC whereas the ACT encoder search is exhaustive.

5.