Adaptive Secondary Transform Sets for Video Coding Beyond AV1

In this paper, a research progress on applying adaptive secondary transform set in the context of ongoing next-generation video coding standardization beyond AV1 (namely AV2) of AOM (Alliance for Open Media) is presented. In the current reference software of AV2, namely AVM (AOM Video Model), secondary transform set is chosen mode-dependently and the index of the selected secondary transform kernel within a given transform set is signaled into the bitstream with the minimum bit cost. With the proposed method, the encoder is empowered with the capability of selecting any transform sets from multiple candidates. Experiment results demonstrate that the proposed method contributes to improving the overall video coding efficiency. The average coding gains measured in weighted YUV PSNR and SSIM BD-Rates are observed as 0.20% and 0.62% respectively for All Intra (AI) configuration of AOM CTC (Common Test Condition) v4 with AVM software of research anchor v4. Coding gains for the higher resolutions 4K and 2K are larger, 0.29% and 0.23% respectively.