Tencent America LLC

Hfi Innovation Inc.

Mediatek Corporation

Tencent Technology (shenzhen) Company Limited

Other

Mediatek Singapore Pte. Ltd.

Xiaozhong Xu

Shan Liu

Xiang Yang Li

Guichun Li

Jun Tian

Xiang Li

Xiang Zhang

Xin Zhao

Chao Huang

Tzu-Der Chuang

Jing Ye

Wei Jiang

Meng Xu

Yu-Chen Sun

Wang-Lin Lai

Wei Wang

Ding Ding

Ching-Yeh Chen

Yu-Wen Huang

Wen Gao

Byeongdoo Choi

Wei W Wang

Thuong Nguyen Canh

Stephan Wenger

Sheng Lin

Lien-Fei Chen

Yiming Li

Jungsun Kim

Zeqiang Li

Liang Zhao

Ding-Yun Chen

Roman Chernyak

Han Huang

Zhenzhong Chen

Yeqing Wu

Weiwei Feng

Yu Wang

Shaw-Min Lei

Jicheng An

Yixin Du

Han Gao

Xianguo Zhang

Yingbin Wang

Chenghao Liu

Ethan Schur

Liqiang Wang

Fang-yi Chao

Chih-Wei Hsu

Anish Tamse

Madhu Peringassery Krishnan

Shih-Ta Hsiang

Ying Hu

Bin Zhu

Biao Wang

Karen Yun Chang

Jacob Yu

Ling Li

Yang Sui

Aspects of the disclosure include methods and apparatuses for video coding. One of the apparatuses includes processing circuitry that receives a bitstream of a current block in a current picture. The current block is coded with a directional nearest neighbor prediction (DNNP) mode. The processing circuitry selects a prediction value for a sample in the current block from a top-left value, a top value, or a left value based on one or more difference values between respective paired values of (i) the top-left value associated with a top-left reference sample that is a top-left neighbor of the current block, (ii) the top value associated with a top reference sample that is a top neighbor of the sample in the current block, and (iii) the left value associated with a left reference sample that is a left neighbor of the sample in the current block. The processing circuitry reconstructs the current block using the selected prediction value for the sample in the current block.

Directional nearest neighbor prediction mode

Processing circuitry performs, based on an input image and a guideline image resulting of an image based computer vision task on the input image, an online training of a compression domain computer vision task framework (CDCVTF). The CDCVTF includes an encoding portion and a decoding portion. The encoding portion compresses the input image into a compression domain, and the decoding portion performs the image based computer vision task in the compression domain. The online training determines at least an update to a neural network in the decoding portion that reduces a loss between a result of the image based computer vision task in the compression domain and the guideline image. The processing circuitry encodes the input image into an encoded image, and forms a bitstream that carries the encoded image and additional bits. The additional bits are indicative of the update to the neural network in the decoding portion.

Online training computer vision task models in compression domain

A method and apparatus that renders a mesh from one or more virtual camera views of an object, the one or more virtual camera views being projections of the mesh onto a plurality of 2D planes; inputs the one or more virtual camera views into a fully-convolutional model for real-time segmentation; generates one or more 2D pixel-wise masks based on the fully-convolutional model, the 2D pixel-wise masks being associated with one or more segmented elements; performs remeshing and mesh optimization using the one or more 2D pixel-wise masks; and generates an output mesh based on the remeshing and mesh optimization.

Mesh optimization using novel segmentation

A method including receiving, at an encoder, a first original polygonal mesh and a second distorted polygonal mesh, the first original polygonal mesh being an original polygonal mesh and the second polygonal mesh being a distorted polygonal mesh; converting the first polygonal mesh and the second polygonal mesh into two or more triangle meshes by subdividing the plurality of polygon faces of each of the first polygonal mesh and the second polygonal mesh into a plurality of triangle faces; sampling a plurality of points on each of the plurality of triangle faces from both the first and the second polygonal meshes; generating at least a first sampled point cloud for one of the first or the second polygonal mesh using the sampled plurality of points; and computing a geometry and attribute distortion profile between the first and the second polygonal meshes based on at least the first sampled point cloud.

Methods of sampling-based objective quality assessment for meshes

Methods and apparatuses for decoding a compressed image using a neural image compression network may be provided. The method may include generating long-range context model parameters associated with a high resolution compressed image, the long-range context model parameters corresponding to a first area. The method may also include splitting the generated long-range context model parameters into a first number of context parameter blocks. The method may also include for each block in the first number of context parameter blocks, predicting respective context features using a long-range context model and respective context parameter blocks, wherein the long-range context model uses a corner-to-center latent decoding strategy or an edge-to-center latent decoding strategy to decode latents associated with the high resolution compressed image. Then, the high resolution compressed image may be reconstructed based on predicted context features.

Block-based long-range context model in neural image compression

Aspects of the disclosure provide methods and apparatuses for mesh coding (encoding and/or decoding). In some examples, an apparatus for coding mesh includes processing circuitry. The processing circuitry decodes, three dimensional (3D) coordinates of vertices in a first 3D mesh frame and connectivity information of the vertices from a bitstream that carries the first 3D mesh frame. The first 3D mesh frame represents a surface of an object with polygons. The processing circuitry deduces texture coordinates associated with the vertices, and decodes a texture map for the first 3D mesh frame from the bitstream. The texture map includes first one or more 2D charts with 2D vertices having the texture coordinates. The processing circuitry reconstructs the first 3D mesh frame based on the 3D coordinates of the vertices, the connectivity information of the vertices, the texture map and the texture coordinates.

Mesh compression with deduced texture coordinates

A method performed by at least one processor of an encoder includes bounding a mesh via a bounding box; partitioning the mesh via a global symmetry plane that partitions the mesh into a first side and a second side that is opposite to the first side; dividing the mesh into a plurality of slices; determining a local symmetry plane for each slice based on the global symmetry plane; based on a determination that a vertex merging condition is satisfied for a first slice from the plurality of slices that shares a boundary with a second slice from the plurality of slices, merging one or more vertices in the boundary of the second slice with one or more vertices in the boundary of the first slice; and encoding each slice from the plurality of slices as a respective sub-mesh of the mesh.

Method and apparatus to encode symmetry mesh with multiple symmetry planes

The various implementations described herein include methods and systems for coding video. In one aspect, a video bitstream includes a current image frame having a current coding block and signals a first syntax element for a residual template cross-component residual model (RT-CCRM) mode. When the RT-CCRM mode is enabled, the computing system identifies, in the current coding block, a first chroma sample and one or more luma samples corresponding to the first chroma sample, determines one or more residuals of the one or more luma samples in the current coding block, and applies a residual filter corresponding to the RT-CCRM mode to generate a first residual of the first chroma sample based on the residuals of the one or more luma samples. The computing system reconstructs the current image frame by compensating a predicted chroma sample with at least the first residual to reconstruct the first chroma sample.

Cross-component residual prediction by using residual template

A method performed by at least one processor in a decoder includes receiving a coded video bitstream that includes a compressed two dimensional (2D) mesh corresponding to a surface of three dimensional (3D) volumetric object. The method includes predicting a sampled 2D coordinate from at least one previously coded sampled 2D coordinate included in the compressed mesh. The method includes deriving a prediction residual associated with the sampled 2D coordinate. The method further includes reconstructing a 2D coordinate corresponding to a boundary vertex based on the predicted sampled 2D coordinate and the derived prediction residual.

Predictive coding of boundary UV information for mesh compression

A method and apparatus comprising computer code configured to cause a processor or processors to obtain an input mesh corresponding to volumetric data of at least one three-dimensional (3D) visual content, derive a curve of a mesh sequence from the input mesh corresponding to volumetric data, the curve including a plurality of vertices of the input mesh of the mesh sequence, subdivide the derived curve of the mesh sequence by adding subdivided vertices, computing a displacement vector for each of the plurality of subdivided vertices; and entropy code the volumetric data based on the displacement vector from at least one of the subdivided vertices.

Mesh vertex displacements coding

A video bitstream is received. The video bitstream includes a current block comprising a plurality of subblocks and a template region of the current block comprising a plurality of template subblocks adjacent to at least one of a top side and a left side of the current block. A motion vector (MV) located in a center position of the current block is determined. The MV is determined based on at least one MV of the plurality of subblocks of the current block. A MV for each of the plurality of template subblocks is determined based on the MV located in the center position of the current block and a respective MV of a corresponding subblock of the plurality of subblocks that is adjacent to the respective template subblock. The current block is reconstructed based on the determined MVs for the plurality of template subblocks.

Motion vector derivation of subblock-based template-matching for subblock based motion vector predictor

A method and an apparatus including processing circuitry are provided. The processing circuitry determines previous probability information associated with each entropy coded region of multiple regions in a previous picture. The processing circuitry selects, based on (i) a location or a quantization parameter of an independently decodable coding segment in a current picture or (ii) syntax information of the independently decodable coding segment, a region among the multiple regions. Initial probability information for one or more current syntax elements of a block in the independently decodable coding segment is determined based on the previous probability information associated with the selected region. The one or more current syntax elements are first to be entropy decoded in the independently decodable coding segment. The processing circuitry entropy decodes coded bits associated with the one or more current syntax elements into a bin string based on the initial probability information.

Adaptive context-based adaptive binary arithmetic coding (CABAC) initial state selection from coded pictures

Aspects of the disclosure provide methods and apparatuses for video encoding/decoding. In some examples, an apparatus for video decoding includes receiving circuitry and processing circuitry. The processing circuitry extracts, from a bitstream, a signal indicative of an application of a template matching based block vector refinement for coding a current block in a current picture referencing (CPR) mode. The processing circuitry derives a first refined block vector associated with the current block according to the template matching based block vector refinement. The first refined block vector includes a first block vector refinement offset applied on a first block vector associated with the current block. The processing circuitry reconstructs the current block according to a first reference block in a same picture as the current block, the first reference block being indicated by the first refined block vector.

Block vector refinement method for intra block copy

The disclosure includes methods and apparatuses that includes processing circuitry that derives a chroma prediction mode of a chroma block in a chroma coding region based on luma coding information of a luma block associated with a position in a reconstructed luma area that is collocated with the chroma block. The chroma coding region is partitioned separately from a collocated luma coding region including the reconstructed luma area. The luma coding information indicates that the luma block is coded with a first coding mode using a first luma BV (an IBC mode or an IntraTMP mode) and a second coding mode. The processing circuitry derives a chroma BV of the chroma block based on a luma BV of the luma block and reconstructs the chroma block based on the chroma prediction mode including the first coding mode and the second coding mode.

Chroma block vector and chroma coding mode derivation

Method, apparatus, and system for texture coordinate prediction for mesh compression are provided. The process may include receiving, for a mesh, a coordinate of a first vertex and a coordinate of a prediction candidate vertex in a three dimensional (3D) space. The process may include determining a stretch perpendicular distance associated with the first vertex and the prediction candidate vertex, the stretch perpendicular distance being based on a conversion of the 3D space into a two dimensional (2D) space, and determining a 2D texture coordinate of the first vertex based on the stretch perpendicular distance associated with the first vertex and the prediction candidate vertex. The process may also include determining a residual of a predicted coordinate of the first vertex and an actual 2D coordinate of the first vertex; and compressing the mesh based on entropy coding the residual.

Texture coordinate prediction in mesh compression

Growing community of inventors

State College, PA, United States of America

Xiaozhong Xu