Alibaba (China) Co., Ltd.

The present disclosure provides a method of encoding a video sequence. The method includes receiving a video sequence; encoding the video sequence using an extrapolation filter-based intra prediction (EIP) fusion mode. The encoding the video sequence using the EIP fusion mode includes obtaining a first predictor by an extrapolation filter-based intra prediction (EIP) mode; obtaining a second predictor by an intra prediction mode; generating a fused predictor by fusing the first predictor and the second predictor; and predicting one or more pictures using the fused predictor.

A video decoding method includes receiving a bitstream; and decoding, using coded information of the bitstream, one or more pictures. The decoding includes decoding a first face feature of a facial image in a first type from the bitstream; translating the first face feature to a second face feature in a second type; and reconstructing the facial image based on the second face feature.

An AVS3 and later-standard encoder and an AVS3 and later-standard decoder are provided, configuring one or more processors of a computing system to perform decoder-side motion vector refinement and affine motion compensation upon a same coding block. The encoder and the decoder are configured to implement one or both of refining a base motion vector of the affine model and refining non-translational parameters of the affine model by searching to minimize bilateral matching (“BM”) cost.

A VVC and later standard encoder and a VVC and later standard decoder are provided, configuring one or more processors of a computing system to perform motion vector refinement for geometric partitioning for motion prediction, and more specifically application of first-pass and second-pass Decoder-side Motion Vector Refinement (“DMVR”) in GPM motion prediction. The VVC and later standard encoder and decoder implement performing motion prediction upon two partitions of a geometric partitioning mode (GPM)-coded coding block, based on respective motion information of each partition, and applying weighted blending to respective motion predictions of each partition of the GPM-coded coding block based on a blending matrix, wherein the blending matrix depends on a split mode.

An AVS3 and later-standard encoder and an AVS3 and later-standard decoder are provided, configuring one or more processors of a computing system to perform motion vector refinement in motion prediction by inter prediction in skip mode or direct mode. The encoder and the decoder are configured to implement preliminary searching before reordering motion candidates of a motion candidate list; prioritized candidate reordering and weighted candidate reordering; an Inter TM candidate list including one or more SBTMVP candidates and ETMVP candidates; TM-based refinement based on combined reference templates; conditional motion vector derivation for TMVP candidates; and converting a bi-predicted block to a uni-predicted block.

A VVC-standard encoder and a VVC-standard decoder are provided, configuring one or more processors of a computing system to apply additional transform cores for inter coded blocks and apply a decoder side intra mode derivation (“DIMD”) process to inter prediction samples or residual samples to derive an intra prediction angular mode.

Embodiments of this application disclose an image classification method, and a method and an apparatus for training an image classification model. A main technical solution includes: obtaining a to-be-classified image; performing feature extraction on the to-be-classified image to obtain a feature representation of the image; performing, by using the feature representation of the image, cross-attention processing on an initial representation of a plurality of cluster centers obtained through pre-training to obtain a plurality of cluster center representations; performing decoding by using the feature representation of the image and the plurality of cluster center representations to obtain a category assignment matrix; and performing classification by using the plurality of cluster center representations and the category assignment matrix to obtain a classification result indicating whether the to-be-classified image belongs to a target category. According to this application, an image classification result can have a higher accuracy rate.

A video decoding method includes: selecting a first intra prediction mode and a second intra prediction mode from a most probable mode (MPM) list; determining a first predictor based on the first intra prediction mode; determining a second predictor based on the second intra prediction mode; blending the first predictor and the second predictor to obtain a blended predictor for intra prediction; and decoding one or more pictures using the blended predictor.

Methods and apparatuses are provided for encoding and decoding video data based on a supplemental enhancement information (SEI) message. An exemplary method includes: generating a reconstructed frame sequence based on a compressed video; decoding a supplemental enhancement information (SEI) message with respect to the reconstructed frame sequence, according to the compressed video; and performing temporal upsampling to the reconstructed frame sequence based on the SEI message by using a neural network.

The present application discloses a computing system and a memory sharing method for a computing system. The computing system includes a first host, a second host, a first memory extension device and a second memory extension device. The first memory extension device is coupled to the first host. The second memory extension device is coupled to the second host and the first memory extension device. The first host accesses a memory space of a memory of the second host through the first memory extension device and the second memory extension device according to a first physical address of the first host.