Google Inc.

Deepmind Technologies Limited

Ching Yin Derek Pang

Yeping Su

Paul Gordon Wilkins

Nishant P Patil

Cheng Che Chen

Chenjie Gu

Akshay Gadde

Christopher Sigurd Fougner

Cheng Chen

Ching-Han Chiang

Hongzi Mao

Marisabel Guevara Hechtman

Daniel James Visentin

Charles Booth Schaff

Alejandro Ramirez Bellido

Jingning Han

Rene Andre Claus

Kyrah Felder

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for encoding video comprising a sequence of video frames. In one aspect, a method comprises for one or more of the video frames: obtaining a feature embedding for the video frame; processing the feature embedding using a rate control machine learning model to generate a respective score for each of multiple quantization parameter values; selecting a quantization parameter value using the scores; determining a cumulative amount of data required to represent: (i) an encoded representation of the video frame and (ii) encoded representations of each preceding video frame; determining, based on the cumulative amount of data, that a feedback control criterion for the video frame is satisfied; updating the selected quantization parameter value; and processing the video frame using an encoding model to generate the encoded representation of the video frame.

Rate control machine learning models with feedback control for video encoding

Methods and systems for encoder parameter setting optimization. A media item to be provided to one or more users of a platform is identified. The media item is associated with a media class. An indication of the identified media item is provided as input to a first machine learning model. The first machine learning model is trained to predict, for a given media item, a set of encoder parameter settings that satisfy a performance criterion in view of a respective media class associated with the given media item. One or more outputs of the first machine learning model are obtained. The one or more obtained outputs include encoder data identifying one or more sets of encoder parameter settings and, for each of the sets of encoder parameter settings, an indication of a level of confidence that a respective set of encoder parameter settings satisfies the performance criterion in view of the media class associated with the identified media item. The identified media item is encoded using the respective set of encoding parameter settings associated with the level of confidence that satisfies a confidence criterion.

Methods and systems for encoder parameter setting optimization

A first aspect is a method for coding a group of pictures (GOP) that includes frames of a video. The method includes encoding, at least some of the frames of the GOP, using a first encoding pass to obtain encoding statistics; obtaining, using the encoding statistics, respective temporal dependency likelihoods (TDLs) for the at least some of the frames of the GOP, where the respective TDLs indicate contributions that the at least some of the frames make in reducing prediction errors of the GOP; obtaining a reference frame based on the respective TDLs; and using the reference frame in encoding at least some of the frames of the GOP in a second encoding pass.

Adaptive GOP structure using temporal dependencies likelihood

Growing community of inventors

San Jose, CA, United States of America

Ching Yin Derek Pang