Abstract: Coding a quantized transform block includes selecting a wavefront scan order for coding quantized transformed coefficients of the quantized transform block. The quantized transform block is of size N×N and the wavefront scan order is such that locations (x, y?1), (x, y?2), and (x, y?3) are sequentially coded and locations (x?1, y), (x?2, y), and (x?3, y) are sequentially coded, for at least one x and one y where 2?x<N and 2?y<N. A probability distribution is selected for coding a quantized transform coefficient of the quantized transform coefficients. A context model for selecting the probability distribution includes at least two immediate neighbors of the quantized transform coefficient in the wavefront scan order. The quantized transformed co-efficient is entropy coded using the probability distribution.

Abstract: A determination is made to use an over-complete transform for decoding a block having N prediction residuals. A primary transform associated with a set of orthonormal primary bases is selected. An additional transform basis is selected. At least N+1 quantized transform coefficients are decoded from a compressed bitstream. The block is then obtained using the at least N+1 quantized transform coefficients, the primary transform, and the additional transform basis.

Abstract: A motion vector for a current block of a current frame is decoded. The motion vector for the current block refers to a first reference block in a first reference frame. A first prediction block of two or more prediction blocks is identified in the first reference frame and using the first reference block. A first grid-aligned block is identified based on the first reference block. A second reference block is identified using a motion vector of the first grid-aligned block in a second reference frame. A second prediction block of the two or more prediction blocks is identified in the second reference frame and using the second reference block. The two or more prediction blocks are combined to obtain a prediction block for the current block.

Abstract: The present disclosure discloses a rubber dome with advanced conduction which includes a support portion, an elastic connection portion, and a pressing portion connected in sequence, the pressing portion having a columnar triggering portion at an axially inner side thereof. An axial end of the triggering portion is spaced at an axial distance b from an axially lower surface of the support portion, and an axial upper surface of the support portion is spaced at a distance h from the axially lower surface of the support portion, and b<h. The pressing portion is axially opened with a bowl-shaped groove extending to the triggering portion. The rubber dome provided by the present disclosure avoids the over-sensitivity phenomenon conventionally caused by lowering the triggering force of the thin film circuit switch by adjusting the conduction distance to make the triggering portion closer to the thin film circuit switch.

Abstract: Decoding using motion vector prediction with derived motion trajectory includes obtaining, from previously reconstructed reference frames available for reconstructing a current frame, reference frame motion fields data for reconstructing the current frame, obtaining, using the reference frame motion fields data, trajectory mapping data for reconstructing the current frame, accessing, from the encoded bitstream, current encoded block data for a current block of the current frame; obtaining a motion vector prediction for the current block in accordance with the trajectory mapping data, obtaining a differential motion vector from the current encoded block data, obtaining a motion vector for the current block by adding the motion vector prediction and the differential motion vector, decoding the current block using the motion vector to obtain decoded block data for the current block, and obtaining reconstructed frame data for the current frame using the decoded block data.

Abstract: A system and method are disclosed. The method includes rendering an image using a three-dimensional (3D) Gaussian splatting process; processing the rendered image with a pretrained latent diffusion model to estimate noise in a latent space; generating a diffusion loss based on a difference between the estimated noise and a sampled noise; periodically applying the diffusion loss to update parameters of the 3D Gaussian splatting process; and generating a novel view synthesis image based on the updated parameters.

Abstract: Decoding a current block of a current frame includes decoding, from a compressed bitstream, one or more syntax elements indicating that a geometric transformation is to be applied; applying the geometric transformation to at least a portion of the current frame to obtain a transformed portion; and obtaining a prediction of the current block based on the transformed portion and an intra-prediction mode.

Abstract: A universal recommendation method based on preference prototype-aware learning, which belongs to the big data analysis technologies. The present disclosure, when realizing the universal cross-domain recommendation, quantitatively learns the user preference through a preference prototype-aware learning method while minimizing interference from the source domain. The method of the present disclosure consists of two complementary components: a hybrid encoder and a preference prototype-aware decoder, which form an end-to-end unified framework suitable for various real-world scenarios. The hybrid encoder uses a hybrid network to learn general representations of interactive items and capture the intrinsic relationships between items across different domains. The preference prototype-aware decoder implements a learnable prototype matching mechanism to quantitatively perceive user preferences and can accurately capture user preferences at a higher semantic level.

Abstract: A cross-domain sequential recommendation method based on time series and projection enhancement. According to the present disclosure, first, single-domain and cross-domain interaction sequences are encoded by using a graph attention mechanism. To account for periodic variation of user preference, timestamp encoding is incorporated to capture temporal characteristics of user behavior. After combining the interaction sequences with the time encoding, a projection mechanism-based module is designed in the present disclosure to accurately capture unique features of a user in a specific domain and shared features in a mixed sequence, such that redundant information in a source domain is effectively prevented from being transmitted to a target domain. Finally, a contrastive learning auxiliary framework is designed to further enhance cross-domain sequence representation.

Abstract: Systems and methods are provided for machine learning models configured as zero-shot personalized text-to-speech models which comprise a feature extractor, a speaker encoder, and a text-to-speech module. The feature extractor is configured to extract acoustic features and prosodic features from new target reference speech associated with the new target speaker. The speaker encoder is configured to generate a speaker embedding corresponding to the new target speaker based on the acoustic features extracted from the new target reference speech. The text-to-speech module is configured to generate the personalized voice corresponding for the new target speaker based on the speaker embedding and the prosodic features extracted from the new target reference speech without applying the text-to-speech module on new labeled training data associated with the new target speaker.

Abstract: Techniques for generating and using a co-located reference frame are described, A first reference frame and a second reference frame for a current frame to be encoded or decoded are reconstructed and used to determine a coarse motion field for the current frame. Fine motion of one or more blocks of the current frame are estimated (e.g., using optical flow estimation). A motion vector of a block of the current frame is updated using the fine motion to result in an updated motion field. A co-located reference frame is determined using the updated motion field, and a prediction process for the current frame is performed using the co-located reference frame. Various techniques may be used to determine whether to adjust the coarse motion field using the fine motion to reduce computing requirements.

Abstract: Spatial motion vector prediction with derived motion trajectory includes determining multiple reference frames available for reconstructing a current frame, determining trajectory mapping data for the current frame that is based on motion fields data between respective frames of the sequence of frames, determining a first reference frame of the multiple reference frames used to encode a neighboring block, determining a second reference frame used to encode the current block, determining a motion vector between the neighboring block and a block position within the second reference frame corresponding to the neighboring block according to the trajectory mapping data and a prediction block location for the neighboring block within the first reference frame, where the motion vector is the motion vector predictor for the current block, and reconstructing the video data of the current block using the motion vector predictor.

Abstract: Decoding using motion vector prediction with derived motion trajectory includes obtaining, from previously reconstructed reference frames available for reconstructing a current frame, reference frame motion fields data for reconstructing the current frame, obtaining, using the reference frame motion fields data, trajectory mapping data for reconstructing the current frame, accessing, from the encoded bitstream, current encoded block data for a current block of the current frame; obtaining a motion vector prediction for the current block in accordance with the trajectory mapping data, obtaining a differential motion vector from the current encoded block data, obtaining a motion vector for the current block by adding the motion vector prediction and the differential motion vector, decoding the current block using the motion vector to obtain decoded block data for the current block, and obtaining reconstructed frame data for the current frame using the decoded block data.

Abstract: A motion vector for a current block of a current frame is decoded. The motion vector for the current block refers to a first reference block in a first reference frame. A first prediction block of two or more prediction blocks is identified in the first reference frame and using the first reference block. A first grid-aligned block is identified based on the first reference block. A second reference block is identified using a motion vector of the first grid-aligned block in a second reference frame. A second prediction block of the two or more prediction blocks is identified in the second reference frame and using the second reference block. The two or more prediction blocks are combined to obtain a prediction block for the current block.

Abstract: Filtering an interpolated reference frame is described. The interpolated reference frame is generated by determining, from a motion field, a motion vector pointing towards a forward reference frame and a motion vector pointing towards a backward reference frame. Expanded prediction blocks, compared to the size of the block of the interpolated reference frame, are determined using the motion vectors and reference frames. The expanded prediction blocks form overlapping areas with adjacent blocks of the interpolated reference frame. The overlapping areas are filtered to mitigate discontinuities.

Abstract: A motion vector for a current block of a current frame is decoded from a compressed bitstream. A location of a reference block within an un-generated reference frame is identified. The reference block is generated using a forward reference frame and a backward reference frame without generating the un-generated reference frame. The reference block is generated by identifying an extended reference block by extending the reference block at each boundary of the reference block by a number of pixels related to a filter length of a filter used in sub-pixel interpolation; and generating pixel values of only the extended reference block by performing a projection using the forward reference frame and the backward reference frame without generating the whole of the un-generated reference frame. The current block is then decoded based on the reference block and the motion vector.

Abstract: Techniques are described for motion vector resolution based motion vector prediction for video coding. A motion vector precision level for coding a current block is determined, a motion vector reference list is generated using the motion vector precision level, an index into the motion vector reference list is determined, where the index identifies a motion vector candidate from the motion vector reference list, and a motion vector for inter prediction of the current block is coded using the motion vector candidate. The motion vector precision level can indicate a single resolution for generating the motion vector reference list or a first resolution for generating the motion vector reference list and a second resolution for coding motion vector residuals of the motion vector.