Abstract: A bitstream that stores encoded image data is described. In addition to the compressed data for color planes of the image, signals identifying respective deblocking filters is identified for the different color planes of the image. The deblocking filters may include those having different lengths for a luma plane as compared to one or more chroma planes of the image. One or more of the color planes, such as the luma plane, may have different filters for filtering reconstructed pixels vertically as compared to filtering the reconstructed pixels horizontally.

Abstract: Mapping-aware coding tools for 360 degree videos adapt conventional video coding tools for 360 degree video data using parameters related to a spherical projection of the 360 degree video data. The mapping-aware coding tools perform motion vector mapping techniques, adaptive motion search pattern techniques, adaptive interpolation filter selection techniques, and adaptive block partitioning techniques. Motion vector mapping includes calculating a motion vector for a pixel of a current block by mapping the location of the pixel within a two-dimensional plane (e.g., video frame) onto a sphere and mapping a predicted location of the pixel on the sphere determined based on rotation parameters back onto the plane. Adaptive motion searching, adaptive interpolation filter selection, and adaptive block partitioning operate according to density distortion based on locations along the sphere.

Abstract: A motion field estimate determined using motion vector information of two or more reference frames of a current/encoded frame is used to derive a motion vector for inter-prediction of the current/encoded frame. Motion trajectory information, including concatenated motion vectors and locations of the current/encoded frame at which those concatenated motion vectors point, is determined by concatenating motion vectors of the reference frames. A motion field estimate is determined using the motion trajectory information and, in some cases, by interpolating unavailable motion vectors using neighbors. The motion field estimate is used to determine a co-located reference frame for the current/encoded frame, and an inter-prediction process is performed for the current/encoded frame using a motion vector derived using the co-located reference frame.

Abstract: Motion prediction using optical flow is determined to be available for a current frame in response to determining that a reference frame buffer includes, with respect to the current frame, a forward reference frame and a backward reference frame. A flag indicating whether a current block is encoded using optical flow is decoded. Responsive to determining that the flag indicates that the current block is encoded using optical flow, a motion vector is decoded for the current block; a location of an optical flow reference block is identified within an optical flow reference frame based on the motion vector; subsequent to identifying the location of the optical flow reference block, the optical flow reference block is generated using the forward reference frame and the backward reference frame without generating the optical flow reference frame; and the current block is decoded based on the optical flow reference block.

Abstract: Techniques for encoding and decoding image data are described. An image is reconstructed and deblocked. A respective deblocking filter is identified for different color planes of the image. The deblocking filters may include those having different lengths for a luma plane as compared to one or more chroma planes of the image. One or more of the color planes, such as the luma plane, may have different filters for filtering reconstructed pixels vertically as compared to filtering the reconstructed pixels horizontally.

Abstract: Decoding a current block of a current frame includes obtaining motion trajectories between the current frame and at least one previously coded frame by projecting motion vectors from the at least one previously coded frame onto the current frame. A motion field is obtained between the current frame and a reference frame used for coding the current frame. The motion field is obtained by extending the motion trajectories from the current frame towards the reference frame. A motion vector for the current block is identified based on the motion field. A prediction block is obtained for the current block using a reference block of the reference frame identified using the motion vector.

Abstract: 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.

Abstract: Video coding using motion prediction coding with coframe motion vectors includes generating a reference coframe spatiotemporally concurrent with a current frame from a sequence of input frames, wherein each frame from the sequence of input frames has a respective sequential location in the sequence of input frames, and wherein the current frame has a current sequential location in the sequence of input frames, generating an encoded frame by encoding the current frame using the reference coframe, including the encoded frame in an encoded bitstream, and outputting the encoded bitstream.

Abstract: Video coding may include generating, by a processor executing instructions stored on a non-transitory computer-readable medium, an encoded frame by encoding a current frame from an input bitstream, by generating a reference coframe spatiotemporally corresponding to the current frame, wherein the current frame is a frame from a sequence of input frames, wherein each frame from the sequence of input frames has a respective sequential location in the sequence of input frames, and wherein the current frame has a current sequential location in the sequence of input frames, and encoding the current frame using the reference coframe. Video coding may include including the encoded frame in an output bitstream and outputting the output bitstream.

Abstract: Mapping-aware coding tools for 360 degree videos adapt conventional video coding tools for 360 degree video data using parameters related to a spherical projection of the 360 degree video data. The mapping-aware coding tools perform motion vector mapping techniques, adaptive motion search pattern techniques, adaptive interpolation filter selection techniques, and adaptive block partitioning techniques. Motion vector mapping includes calculating a motion vector for a pixel of a current block by mapping the location of the pixel within a two-dimensional plane (e.g., video frame) onto a sphere and mapping a predicted location of the pixel on the sphere determined based on rotation parameters back onto the plane. Adaptive motion searching, adaptive interpolation filter selection, and adaptive block partitioning operate according to density distortion based on locations along the sphere.

Abstract: Dynamic motion vector referencing is used to predict motion within video blocks. A motion trajectory is determined for a current frame including a video block to encode or decode based on a reference motion vector used for encoding or decoding one or more reference frames of the current frame. One or more temporal motion vector candidates are then determined for predicting motion within the video block based on the motion trajectory. A motion vector is selected from a motion vector candidate list including the one or more temporal motion vector candidates and used to generate a prediction block. The prediction block is then used to encode or decode the video block. The motion trajectory is based on an order of video frames indicated by frame offset values encoded to a bitstream. The motion vector candidate list may include one or more spatial motion vector candidates.

Abstract: Entropy coding a sequence of symbols is described. A first probability model for entropy coding is selected. At least one symbol of the sequence is coded using a probability determined using the first probability model. The probability according to the first probability model is updated with an estimation of a second probability model to entropy code a subsequent symbol. The combination may be a fixed or adaptive combination.

Abstract: A transform type is obtained for decoding the transform block of transform coefficients. A template for entropy-decoding values related to the transform coefficients is selected based on the transform type. The template indicates, for a to-be-coded value, positions of already coded values. A context for selecting a probability distribution for entropy decoding a current value of the values is determined using the template. The current value is entropy decoded from a compressed bitstream using the probability distribution.

Abstract: Decoding a current block of a current frame includes obtaining motion trajectories between the current frame and at least one previously coded frame by projecting motion vectors from the at least one previously coded frame onto the current frame. A motion field is obtained between the current frame and a reference frame used for coding the current frame. The motion field is obtained by extending the motion trajectories from the current frame towards the reference frame. A motion vector for the current block is identified based on the motion field. A prediction block is obtained for the current block using a reference block of the reference frame identified using the motion vector.

Abstract: Transform kernel candidates including a vertical transform type associated with a vertical motion and a horizontal transform type associated with a horizontal motion can be encoded or decoded. During a decoding operation, a probability model for decoding encoded bitstream video data associated with a transform kernel candidate for an encoded transform block is identified based on one or both of a first transform kernel candidate selected for an above neighbor transform block of the encoded transform block or a second transform kernel candidate selected for a left neighbor transform block of the encoded transform block. The encoded bitstream video data associated with the transform kernel candidate is decoded using the probability model.

Abstract: Motion prediction using optical flow is determined to be available for a current frame in response to determining that a reference frame buffer includes, with respect to the current frame, a forward reference frame and a backward reference frame. A flag indicating whether a current block is encoded using optical flow is decoded. Responsive to determining that the flag indicates that the current block is encoded using optical flow, a motion vector is decoded for the current block; a location of an optical flow reference block is identified within an optical flow reference frame based on the motion vector; subsequent to identifying the location of the optical flow reference block, the optical flow reference block is generated using the forward reference frame and the backward reference frame without generating the optical flow reference frame; and the current block is decoded based on the optical flow reference block.

Abstract: Transform kernel candidates including a vertical transform type associated with a vertical motion and a horizontal transform type associated with a horizontal motion can be encoded or decoded. During an encoding operation, a residual block of a current block is transformed according to a selected transform kernel candidate to produce a transform block. A probability model for encoding the selected transform kernel candidate is then identified based on neighbor transform blocks of the transform block. The selected transform kernel candidate is then encoded according to the probability model. During a decoding operation, the encoded transform kernel candidate is decoded using the probability model. The encoded transform block is then decoded by inverse transforming dequantized transform coefficients thereof according to the decoded transform kernel candidate.

Abstract: Decoding a current frame includes identifying a first reference frame and a second reference frame for decoding the current frame; storing reference motion vectors of reference blocks of the first reference frame, where other reference frames are used to decode the first reference frame; identifying motion trajectories that pass through the current frame by projecting the reference motion vectors of the reference blocks of the first reference frame onto the current frame using at least a third reference frame of the other reference frames, where the projecting identifies, for a first current block of the current frame a corresponding first reference block in the first reference frame, and a corresponding reference motion vector of the reference motion vectors is associated with the corresponding first reference block; and projecting the corresponding reference motion vector onto the second reference frame to obtain a second reference block in the second reference frame.

Abstract: A processor is configured to maintain, for encoding current values related to the transform coefficients a first line buffer and a second line buffer. The current values are arranged along a current scan-order anti-diagonal line. The first line buffer includes first values of a first scan-order anti-diagonal line. The second line buffer includes second values of a second scan-order anti-diagonal line. The processor is further configured to interleave the first values and the second values in a destination buffer; select, using the destination buffer, a probability distribution for coding a current value of the current values; entropy encode, in a compressed bitstream, the current value using the probability distribution; and replace, for coding values of an immediately subsequent scan-order anti-diagonal line to the current scan-order anti-diagonal line, one of the second line buffer or the first line buffer with the current scan-order anti-diagonal line.

Abstract: Decoding a video stream may include decoding a first block of a current frame by decoding a first motion vector from the encoded video stream, decoding an identifier of a first interpolation filter from the encoded video stream, and reconstructing the first block using the first motion vector and the first interpolation filter. Decoding a second block of the current frame may include identifying the first motion vector from the first block as a selected motion vector for predicting the second block in response to decoding an inter-prediction mode identifier for decoding the second block, identifying the first interpolation filter as a selected interpolation filter for predicting the second block in response to identifying the first motion vector from the first block as the selected motion vector for predicting the second block, and reconstructing the second block using the first motion vector and the first interpolation filter.