Embodiments described herein provided for an instruction and associated logic to enable a processing resource including a tensor accelerator to perform optimized computation of sparse submatrix operations. One embodiment provides hardware logic to apply a numerical transform to matrix data to increase the sparsity of the data. Increasing the sparsity may result in a higher compression ratio when the matrix data is compressed.

A method, apparatus, and computer-readable medium for point cloud coefficient coding are provided. The method may include receiving compressed point cloud data based on set-index values and symbol-index values; and entropy-decoding the set-index values based on the compressed point cloud data. The symbol-index values may be bypass-decoded based on the compressed point cloud data, and the set-index values and symbol-index values may be combined into transform coefficients associated with cloud point data.

In some examples, a method for compressing a spectral reflectance dataset may be performed through compression circuitry. The method may include computing a principal component analysis basis for the spectral reflectance dataset; projecting the spectral reflectance dataset onto the principal component analysis basis to obtain a weight matrix; quantizing the weight matrix; performing a Huffman encoding process on the quantized weight matrix to generate a Huffman table and Huffman codes for the quantized weight matrix; and providing compressed spectral reflectance data as the principal component analysis basis, the Huffman table, and the Huffman codes.

An image coding device including: an edge detecting section configured to perform edge detection using an image signal of a reference image for a coding object block; a transform block setting section configured to set transform blocks by dividing the coding object block such that a boundary between the blocks after division does not include an edge on a basis of a result of the edge detection; and a coding processing section configured to generate coded data by performing processing including an orthogonal transform of each of the transform blocks.

For one embodiment of the present invention, a method of object instance edge detection and segmentation is described. The method includes obtaining an input image with a shape and extracting, with a feature extractor of a point supervised transformer model, a hierarchical combination of features from the input image including a set of feature maps having different levels. The method further includes receiving, with a transformer decoder, an output including a feature map from the feature extractor, and object queries each with d dimensions, training the point supervised transformer model with a sparse set of keypoint annotations along a boundary of each object instance, and generating a box prediction, a classification prediction, and a coefficient prediction for each object instance based on an output from the transformer decoder.

An encoding method includes; acquiring a first image; separating the first image into a plurality of second images by extracting a pixel in the first image after every predetermined number of pixels in each of horizontal and vertical directions of the first image; and encoding each of the separated second images. By transmitting those pieces of encoded data, even if a packet loss occurs in one of the second images, the missing pixel can be re-generated based on corresponding neighboring pixels in other second images.

There is provided a computerized method and system of controlling a quality measure in a compression quality evaluation system, the method comprising: calculating a grain value indicative of an extent of grain present in an input image, the grain value being calculated based on one or more features characterizing a base image related to the input image; and configuring the quality measure upon a criterion being met by the value, the quality measure being indicative of perceptual quality of a compressed image compressed from the input image. The calculated grain value may be dependent also on further characteristics of the input image, or in the case of a sequence of images, dependent also on the relation between the image and the preceding image.

A video signal processing method may comprise the steps of: confirming a prediction mode applied to a current coding unit; confirming whether a plurality of preset conditions are satisfied on the basis of at least one of the prediction mode of the current coding unit and a size of the current coding unit; parsing a first syntax element indicating a transform kernel applied to a transform unit included in the current coding unit when the plurality of preset conditions are satisfied; determining a transform kernel applied to horizontal and vertical directions of the current transform unit on the basis of the first syntax element; and generating a residual signal of the current transform unit by performing an inverse transform on the current transform unit using the determined transform kernel.

The present invention relates to an improved method and apparatus for image compression and particularly to an improved block coding apparatus and method for compression for use with the JPEG2000 standard, although not limited to this. Methods for coding and decoding blocks and subbands samples derived from still images video frames or related media, involving three bit-streams and the partitioning of samples from the blocking to define groups, is provided. A first bit-stream encodes the significance of whole groups. A second bit-stream encodes the significance of individual samples within each group. The second bit-stream also encodes an unsigned residual value for each significant group. A third bit stream provides a sign bit and any additional magnitude bits required to represent the significant sample values. Exponent predictors are computal using both exponent bounds and the additional magnitude bits associated with previous samples in the block.

Methods of performing a complex Fourier transform of a complex data set corresponding to an image are disclosed. The methods comprise receiving a complex data set and performing a first 1D complex Fourier transform in the complex data set in Cartesian form; converting the complex data set into polar form and compressing the complex data set in polar form; performing a row-column transformation of the complex data set; decompressing the complex data set and converting the complex data set back into Cartesian form; and performing a second 1D Fourier transform in the complex data set in Cartesian form, wherein the second 1D complex Fourier transform is orthogonal to the first 1D complex Fourier transform. Corresponding systems are also disclosed, as are application to the iterative computation of computer-generated holograms.