JVET-AK0083 AHG15: Analysis of high-level control of intra prediction methods in gaming content [X. Liang, K. Choi (Kyung Hee Univ.), S. Lee (Atins)]

This contribution investigates the high-level control of intra prediction methods for game content. In ECM-15.0, the approach of high-level control of intra prediction methods has been adopted in the ECM software, demonstrating efficient improvements in coding performance for screen content. In this contribution, the method was tested to analyze its behavior with game content. The impact on BD-rate performance when applying the high-level control of intra prediction methods under the ECM CTC for game applications is reported as follows:

Test1 (whole tools off):
AI: { 0.11% Y, 0.04% U, 0.04% V, 98.2% EncT, 89.0% DecT } for SDR Class (G1, G2)
AI: { 0.20% Y, 0.12% U, 0.30% V, 99.0% EncT, 102.7% DecT } for HDR Class (G3)

Test2 (adaptive tools off):
AI: { 0.05% Y, 0.03% U, 0.03% V, 99.5% EncT, 100.6% DecT } for SDR Class (G1, G2)
AI: { 0.00% Y, 0.00% U, 0.00% V, 92.8% EncT, 95.6% DecT } for HDR Class (G3)

The complete or adaptive disabling of three tools (reference sample filtering, PDPC, intra prediction fusion) was originally introduced to avoid degrading of graphics-generated content. In CTC of other classes, tools are never turned off for camera content, and adaptively turned off for classes F and TGM.

For gaming content CTC also uses the three tools always enabled. The results indicate that PSNR-wise, this may be correct, however examples are shown where visual quality is better with turning tools.

Further study would be needed to investigate this. As gaming content is sometimes similar to regular video, if at all, an adaptive disabling would be appropriate, or potentially more specific encoder optimization would be necessary.