公开(公告)号：US10674152B2

公开(公告)日：2020-06-02

申请号：US16134134

申请日：2018-09-18

Applicant: GOOGLE LLC

Inventor： Claudionor Coelho ,  Dake He ,  Aki Kuusela ,  Shan Li

IPC: H04N19/124 ,  H04N19/164 ,  H04N19/96 ,  H04N19/176

Abstract: A method for encoding an image block includes presenting, to a machine-learning model, the image block and a first value corresponding to a first quantization parameter; obtaining first mode decision parameters from the machine-learning model; and encoding the image block using the first mode decision parameters. The first value results from a non-linear function using the first quantization parameter as input. The machine-learning model is trained to output mode decision parameters by using training data. Each training datum includes a training block that is encoded by a second encoder, second mode decision parameters used by the second encoder for encoding the training block, and a second value corresponding to a second quantization parameter. The second encoder used the second quantization parameter for encoding the training block and the second value results from the non-linear function using the second quantization parameter as input.

公开(公告)号：US20200084478A1

公开(公告)日：2020-03-12

申请号：US16127637

申请日：2018-09-11

Applicant: GOOGLE LLC

Inventor： Ruijie Xu ,  Dake He

IPC: H04N19/91 ,  H04N19/70 ,  H04N19/13 ,  H04N19/124

Abstract: Separate pathways for filtering and for machine learning are introduced within a video coder. A first pathway filters a first copy of a reconstructed frame to produce a filtered frame, which is included in an output video stream for display at a user device. A second pathway processes a second copy of the reconstructed frame using a learning model, such as for training and/or inference. The first and second pathways are introduced after the reconstruction stage of an encoder or decoder. The input to each of the first and second pathways is thus produced without using a non-injective function, and, while the first pathway includes at least one non-injective function, the second pathway does not. As a result, training the learning model using the second copy of the reconstructed frame results in a greater classification accuracy upper bound than training the learning model using the filtered frame.

公开(公告)号：US20200029098A1

公开(公告)日：2020-01-23

申请号：US16042261

申请日：2018-07-23

Applicant: GOOGLE LLC

Inventor： Jingning Han ,  Aki Kuusela ,  Stanislav Vitvitskyy ,  Yaowu Xu ,  Ching-Han Chiang ,  Dake He

IPC: H04N19/91 ,  H04N19/70 ,  H04N19/61 ,  H04N19/44

Abstract: Asymmetric probability model updating and entropy coding includes using different numbers of bits for storing probabilities of a probability model and for entropy coding symbols using that probability model. The probabilities of a probability model are updated according to values of syntax elements decoded from a bitstream. The probabilities are associated with possible values of the syntax elements and are stored using a first bit precision. Based on the updated probabilities, a second bit precision to use to entropy decode the syntax elements is determined. The second bit precision is less than the first bit precision. The syntax elements are then entropy decoded using the second bit precision, such as to produce quantized transform coefficients, which may be further processed and output to an output video stream. Using the first bit precision to entropy decode the syntax elements results in a lower compression throughput than using the second bit precision.

公开(公告)号：US10484695B2

公开(公告)日：2019-11-19

申请号：US15798495

申请日：2017-10-31

Applicant: GOOGLE LLC

Inventor： Joseph Young ,  Dake He

IPC: H04N19/159 ,  H04N19/176 ,  H04N11/02 ,  H04N19/129 ,  H04N19/18 ,  H04N19/60 ,  H04N19/13 ,  H04N19/124 ,  H04N19/44 ,  H04N19/91 ,  H04N19/122

Abstract: Coding using level maps is disclosed. A method includes coding a scan position, in a forward scan direction, corresponding to an end-of-block and coding, in a backward scan direction, a non-zero map indicating positions of the transform block containing non-zero transform coefficients. The method also includes coding, in the backward scan direction, lower-range level maps, each lower-range level map having a respective map level up to a maximum map level, the lower-range level map indicating which absolute values of the non-zero transform coefficients are equal to the respective map level and which absolute values of the non-zero transform coefficients are greater than the respective map level. The method also includes coding a coefficient residual map, each residual coefficient of the coefficient residual map corresponding to a respective non-zero transform coefficient of the transform block having an absolute value exceeding the maximum map level.

公开(公告)号：US10448019B2

公开(公告)日：2019-10-15

申请号：US15824058

申请日：2017-11-28

Applicant: GOOGLE LLC

Inventor： Dake He

IPC: H04N19/13 ,  H04N19/18 ,  H04N19/159 ,  H04N19/196 ,  H04N19/124 ,  H04N19/176 ,  H04N19/91 ,  H04N19/96 ,  H04N19/129

Abstract: Entropy encoding and decoding a sequence of symbols using probability mixing is disclosed. A method includes selecting models that include a first model and a second model, for at least a symbol, at a position of the symbols, determining a mixed probability using the first model and the second model, and coding the symbol using the mixed probability. Determining the mixed probability for the symbol includes determining, using the first model, a first conditional probability for coding the symbol, determining, using the second model, a second conditional probability for coding the symbol, and determining, using the first conditional probability and the second conditional probability, the mixed probability for coding the symbol. The first conditional probability is a conditional probability of the symbol given a sub-sequence of the sequence up to the position. The second conditional probability being a conditional probability of the symbol given the sub-sequence.

公开(公告)号：US10382758B2

公开(公告)日：2019-08-13

申请号：US15880939

申请日：2018-01-26

Applicant: GOOGLE LLC

Inventor： Jingning Han ,  Dake He

IPC: H04N7/12 ,  H04N11/02 ,  H04N11/04 ,  H04N19/129 ,  H04N19/61 ,  H04N19/82 ,  H04N19/176 ,  H04N19/18

Abstract: A scan order for encoding or decoding coefficients of a transform block is selected on a transform block-level. A set of candidate scan orders is processed by identifying end of block positions within the transform block for each of the candidate scan orders. Cost values are determined for each of the candidate scan orders to reflect a number of the coefficients of the transform block that are located before the respective end of block positions. In particular, a cost value for a candidate scan order reflects the number of zero-value coefficients located before the end of block position for that candidate scan order. One of the candidate scan orders is then selected based on those cost values. The selected scan order is used to scan the coefficients in the transform block, such as for encoding those coefficients to a bitstream or for decoding those coefficients to an output video stream.

公开(公告)号：US20190058889A1

公开(公告)日：2019-02-21

申请号：US15681493

申请日：2017-08-21

Applicant: GOOGLE LLC

Inventor： Dake He

IPC: H04N19/18 ,  H04N19/129 ,  H04N19/467 ,  H04N19/61

Abstract: A method for decoding a transform block of quantized transform coefficients includes decoding a predetermined number of coefficients of the quantized transform coefficients, determining a value for the predetermined number of coefficients, and decoding a subsequent quantized transform coefficient by reading bits from the encoded bitstream and traversing a coefficient token tree having a root node indicating an EOB token. The decoding of the subsequent quantized transform coefficient uses the value to determine whether to traverse the coefficient token tree starting at the root node or at another node. A method for encoding a transform block of quantized transform coefficients includes partitioning the quantized transform coefficients into at least a first coefficients group and a second coefficients group, determining a value of the first coefficients group, and encoding, based on the value, a bit indicative of an end-of-block (EOB) for a transform coefficient of the second coefficients group.

公开(公告)号：US20190045225A1

公开(公告)日：2019-02-07

申请号：US15668745

申请日：2017-08-04

Applicant: GOOGLE LLC

Inventor： Ruijie Xu ,  Dake He

IPC: H04N19/91 ,  H04N19/44 ,  H04N19/60 ,  H04N19/124

Abstract: Systems and methods are disclosed for entropy coding of blocks of image data. For example, methods may include partitioning a block of video data into a plurality of groups of elements; decoding, using an entropy decoder, data from an encoded bitstream to obtain elements of a first group from the plurality of groups of elements; determining a category based on the elements of the first group; based on the category, selecting a context for an element from a second group from the plurality of groups of elements; and decoding, using the entropy decoder using the context, data from the encoded bitstream to obtain the element of the second group from the plurality of groups of elements.