IP Library Granted Patent US 12689728
Granted Patent B2
US 12689728 · App. 18/776,760 · Granted Jul 21, 2026

Neural network based transform set selection

Inventors: Xin Zhao (Santa Clara, CA); Madhu Peringassery Krishnan (Mountain View, CA); Shan Liu (San Jose, CA)
Assignee: TENCENT AMERICA LLC
H04N19/12H04N19/105H04N19/159H04N19/176H04N19/18
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12689728
App. No.
18/776,760
Granted
Jul 21, 2026
Kind
B2
Abstract

Systems and methods for coding and decoding of a coded bitstream is provided. A method includes encoding a block of a picture. The encoding includes selecting a transform set based on at least one neighboring sample from one or more previously encoded neighboring blocks or from a previously encoded picture and transforming coefficients of the block using a transform from the transform set.

Claims (33)

1 . A method performed by at least one processor, the method comprising:

decoding a block of a picture, the decoding comprising:

selecting a transform set based on at least one neighboring sample from one or more neighboring blocks or from a picture, and

the at least one neighboring sample is an input to a neural network whose output is a selection of the transform set and an identifier of a prediction mode set; and

transforming coefficients of the block using a transform from the transform set, and the selecting the transform set comprises:

selecting a sub-group of transform sets from a group of transform sets based on information of an intra prediction mode or an inter prediction mode; and

selecting the transform set from the sub-group.

2 . The method of claim 1 , wherein the neural network identifies the transform set as for at least one of a secondary transform, a primary transform, and a combination of the secondary transform and the secondary transform.

3 . The method of claim 2 , wherein the secondary transform indicates a non-separable transform scheme, and wherein the primary transform indicates any of different types of discrete cosine transforms, discrete sine transforms, line graph transforms with respectively different self-loop rates.

4 . The method of claim 1 , wherein the at least one neighboring sample is both at least one of upsampled and downsampled and also an input to the neural network.

5 . The method of claim 1 , wherein the at least one neighboring sample is scaled and input to the neural network.

6 . The method of claim 1 , wherein parameters of the neural network are based on any of whether the block is intra coded, a block width, a block height, a quantization parameter, whether a current picture is coded as an intra frame, and an intra prediction mode.

7 . The method of claim 1 , wherein the at least one neighboring sample includes one or more lines of top and left neighboring reconstructed samples.

8 . A method performed by at least one processor, the method comprising: encoding a block of a picture, the encoding comprising:

selecting a transform set based on at least one neighboring sample from one or more neighboring blocks or from a picture, and the at least one neighboring sample is an input to a neural network whose output is a selection of the transform set and an identifier of a prediction mode set; and

transforming coefficients of the block using a transform from the transform set, and the selecting the transform set comprises:

selecting a sub-group of transform sets from a group of transform sets based on information of an intra prediction mode or an inter prediction mode; and

selecting the transform set from the sub-group.

9 . The method of claim 8 , wherein the neural network identifies the transform set as for at least one of a secondary transform, a primary transform, and a combination of the secondary transform and the secondary transform.

10 . The method of claim 9 , wherein the secondary transform indicates a non-separable transform scheme, and wherein the primary transform indicates any of different types of discrete cosine transforms, discrete sine transforms, line graph transforms with respectively different self-loop rates.

11 . The method of claim 8 , wherein the at least one neighboring sample is both at least one of upsampled and downsampled and also an input to the neural network.

12 . The method of claim 8 , wherein the at least one neighboring sample is scaled and input to the neural network.

13 . The method of claim 8 , wherein parameters of the neural network are based on any of whether the block is intra coded, a block width, a block height, a quantization parameter, whether a current picture is coded as an intra frame, and an intra prediction mode.

14 . The method of claim 8 , wherein the at least one neighboring sample includes one or more lines of top and left neighboring reconstructed samples.

15 . A non-transitory computer-readable storage medium storing a video bitstream that is generated by a video encoding method, the video encoding method comprising: encoding and transmitting a block of a picture, the encoding comprising:

selecting a transform set based on at least one neighboring sample from one or more neighboring blocks or from a picture, and the at least one neighboring sample is an input to a neural network whose output is a selection of the transform set and an identifier of a prediction mode set; and

transforming coefficients of the block using a transform from the transform set, and the selecting the transform set comprises:

selecting a sub-group of transform sets from a group of transform sets based on information of an intra prediction mode or an inter prediction mode; and

selecting the transform set from the sub-group.

16 . The method of claim 15 , wherein the neural network identifies the transform set as for at least one of a secondary transform, a primary transform, and a combination of the secondary transform and the secondary transform.

17 . The method of claim 16 , wherein the secondary transform indicates a non-separable transform scheme, and wherein the primary transform indicates any of different types of discrete cosine transforms, discrete sine transforms, line graph transforms with respectively different self-loop rates.

18 . The method of claim 15 , wherein the at least one neighboring sample is both at least one of upsampled and downsampled and also an input to the neural network.

19 . The method of claim 15 , wherein the at least one neighboring sample is scaled and input to the neural network.