Back to Search Document details
35th Meeting: Sapporo, JP, July 2024 2024-07-15 07:16
Non-EE2: Regression-based SGPM blending
Abstract
This contribution proposes to extend the SGPM candidate list with regression-based SGPM candidates. Each regression-based SGPM candidate contains two intra prediction modes and a derived blending matrix. The two intra prediction modes are selected from the primary MPM list and up to 6 block vectors. The matrix-based intra prediction replacement method is also extended to SGPM. It is reported that on top of ECM-13.0, the overall coding performance impact for {Y, U, V, EncT, DecT} is {-0.12%, -0.01%, -0.02%, 103.9%, 102.3%} for AI and {-%, -%, -%, %, %} for RA.
Citation