US 12,143,605 B2
Inter-prediction mode-dependent transforms for video coding
Rohit Singh, Sunnyvale, CA (US); Debargha Mukherjee, Cupertino, CA (US); Elliott Karpilovsky, Santa Clara, CA (US); and Lester Lu, Los Angeles, CA (US)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Dec. 6, 2021, as Appl. No. 17/543,126.
Application 17/543,126 is a continuation of application No. 16/919,507, filed on Jul. 2, 2020, granted, now 11,197,004.
Prior Publication US 2022/0094950 A1, Mar. 24, 2022
Int. Cl. H04N 19/184 (2014.01); H04N 19/124 (2014.01); H04N 19/61 (2014.01)
CPC H04N 19/184 (2014.11) [H04N 19/124 (2014.11); H04N 19/61 (2014.11)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
generating a prediction residual for a current video block using a reference frame;
determining, based on side information usable to encode one or both of the current video block or the reference frame and including data other than pixel values of the one or both of the current video block or the reference frame, a trained transform for transforming the prediction residual from amongst multiple trained transforms determined using individual side information types and combinations of the individual side information types, wherein the side information represents values of one of the individual side information types or one of the combinations of the individual side information types, wherein the multiple trained transforms are determined based on statistical differences between video blocks of a training data set resulting from encoding or decoding the video blocks, and wherein each of the multiple trained transforms includes a different pair of horizontal and vertical transforms;
transforming the prediction residual according to the trained transform; and
encoding the side information and data associated with the transformed prediction residual to a bitstream.