Abstract: The performance of a neural network is improved by applying quantization to data at various points in the network. In an embodiment, a neural network includes two paths. A quantization is applied to each path, such that when an output from each path is combined, further quantization is not required. In an embodiment, the neural network is an autoencoder that includes at least one skip connection. In an embodiment, the system determines a set of quantization parameters based on the characteristics of the data in the primary path and in the skip connection, such that both network paths produce output data in the same fixed point format. As a result, the data from both network paths can be combined without requiring an additional quantization.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: The performance of a neural network is improved by applying quantization to data at various points in the network. In an embodiment, a neural network includes two paths. A quantization is applied to each path, such that when an output from each path is combined, further quantization is not required. In an embodiment, the neural network is an autoencoder that includes at least one skip connection. In an embodiment, the system determines a set of quantization parameters based on the characteristics of the data in the primary path and in the skip connection, such that both network paths produce output data in the same fixed point format. As a result, the data from both network paths can be combined without requiring an additional quantization.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: The performance of a neural network is improved by applying quantization to data at various points in the network. In an embodiment, a neural network includes two paths. A quantization is applied to each path, such that when an output from each path is combined, further quantization is not required. In an embodiment, the neural network is an autoencoder that includes at least one skip connection. In an embodiment, the system determines a set of quantization parameters based on the characteristics of the data in the primary path and in the skip connection, such that both network paths produce output data in the same fixed point format. As a result, the data from both network paths can be combined without requiring an additional quantization.

Abstract: Disclosed approaches may leverage the actual spatial and reflective properties of a virtual environment—such as the size, shape, and orientation of a bidirectional reflectance distribution function (BRDF) lobe of a light path and its position relative to a reflection surface, a virtual screen, and a virtual camera—to produce, for a pixel, an anisotropic kernel filter having dimensions and weights that accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface. In order to accomplish this, geometry may be computed that corresponds to a projection of a reflection of the BRDF lobe below the surface along a view vector to the pixel. Using this approach, the dimensions of the anisotropic filter kernel may correspond to the BRDF lobe to accurately reflect the spatial characteristics of the virtual environment as well as the reflective properties of the surface.

Abstract: A method for improving performance of generation of digitally represented graphics. The method comprises: receiving a first representation of a base primitive; providing a set of instructions associated with vertex position determination; executing said retrieved set of instructions on said first representation of said base primitive using bounded arithmetic for providing a second representation of said base primitive, and subjecting said second representation of said base primitive to a culling process. A corresponding apparatus and computer program product are also presented.

Abstract: A unified compression/decompression architecture is disclosed for reducing memory bandwidth requirements in 3D graphics processing applications. The techniques described erase several distinctions between a texture (compressed once, and decompressed many times), and buffers (compressed and decompressed repeatedly during rendering of an image). An exemplary method for processing graphics data according to one or more embodiments of the invention thus begins with the updating of one or more tiles of a first image array, which are then compressed, using a real-time buffer compression algorithm, to obtain compressed image array tiles. The compressed image array tiles are stored for subsequent use as a texture. During real-time rendering of a second image array, the compressed image array tiles are retrieved and decompressed using a decompression algorithm corresponding to the buffer compression algorithm.

Abstract: A computer graphics processor (20,50) and a method for rendering a three-dimensional image on a display screen. The computer graphics processor (20,50) comprises a rasterizer (23,53) configured to perform pixel traversal of a primitive after projection of the primitive. Furthermore, the rasterizer (23,53) is configured to perform the pixel traversal of a first primitive for a plurality of views prior to performing pixel traversal of a next primitive for one or several views.

Abstract: This relates to a generation of digitally represented graphics. A first representation of a group of vertices is received. A second representation of said group of vertices is determined based on said first representation. A first set of instructions is executed on said second representation of said group of vertices for providing a third representation of said group of vertices, said first set of instructions being associated with vertex position determination. The third representation of said group of vertices is subjected to a culling process.

Abstract: First and second codewords are determined, based on first feature vector components of the image elements in an image block, as representations of a first and second component value. Third and fourth codewords are determined, based on second vector components, as representations of a third and fourth component value. First N1 and second N2 resolution numbers are selected based on the relation of a distribution of the first vector components and a distribution of the second vector components. N1 additional component values are generated based on the first and second component values and N2 additional component values are generated based on the third and fourth component values. Component indices indicative of the generated component values are then provided for the different image elements.

Abstract: First and second codewords are determined, based on first feature vector components of the image elements in an image block, as representations of a first and second component value. Third and fourth codewords are determined, based on second vector components, as representations of a third and fourth component value. First N1 and second N2 resolution numbers are selected based on the relation of a distribution of the first vector components and a distribution of the second vector components. N1 additional component values are generated based on the first and second component values and N2 additional component values are generated based on the third and fourth component values. Component indices indicative of the generated component values are then provided for the different image elements.

Abstract: A block (300) of image elements (310) is compressed by determining multiple base vectors (510, 520, 530, 540) based on the feature vectors (312) associated with the image elements. Additional vectors (560, 570) are calculated based on defined pairs of neighboring base vectors (510, 520, 530, 540). A vector among the base vectors (510, 520, 530, 540) and the additional vectors (560, 570) is selected as representation of the feature vector (312) of an image element (310). An identifier (550) associated with selected vector is assigned to the image element (310) and included in the compressed block (500) which also comprises representations of the determined base vectors (510, 520, 530, 540).

Abstract: First and second codewords are determined, based on first feature vector components of the image elements in an image block, as representations of a first and second component value. Third and fourth codewords are determined, based on second vector components, as representations of a third and fourth component value. First N1 and second N2 resolution numbers are selected based on the relation of a distribution of the first vector components and a distribution of the second vector components. N1 additional component values are generated based on the first and second component values and N2 additional component values are generated based on the third and fourth component values. Component indices indicative of the generated component values are then provided for the different image elements.

Abstract: A block (300) of image elements (310) is compressed by identifying a base vector (460) based on normalized feature vectors (312) of the block (300). If a position-determining coordinate (420) of the base vector (460) is present inside a defined selection section (530) of feature vector space (500), the block (300) is compressed according to a default mode and an auxiliary mode to get a default and auxiliary compressed block (600), respectively. The compressed block (600) resulting in smallest compression error is selected. If the auxiliary mode is selected, the position-determining coordinate (420) is mapped to get a mapped coordinate (425) present outside the representable normalization portion (510) of vector space (500). The auxiliary compressed block (600) comprises a representation of this mapped coordinate (425). If the default mode is selected no such coordinate mapping is performed and the default compressed block (600) instead comprises a representation of the non-mirrored coordinate (420).

Abstract: Hierarchical bounding of displaced parametric surfaces may be a very common use case for tessellation in interactive and real-time rendering. An efficient normal bounding technique may be used, together with min-max mipmap hierarchies and oriented bounding boxes. This provides substantially faster convergence for the bounding volumes of the displaced surface, without tessellating and displacing the surface in some embodiments. This bounding technique can be used for different types of culling, ray tracing, and to sort higher order primitives in tiling architectures.