Back to results
US 2026/0311459 A1Published application
4:2:0 packing of feature maps
AssigneeCanon Kabushiki Kaisha
InventorChristopher James Rosewarne
Life of this patent
Estimating the term…
Abstract
A method of decoding feature maps from encoded data. A plurality of samples is decoded from the encoded data. The feature maps are determined based on one image from at least a first group of samples arranged in a first two-dimensional array and a second group of samples arranged in a second two-dimensional array, where the second two-dimensional array is different from the first two-dimensional array.
Claims5
A decoding method comprising:
- decoding, from a bitstream, a syntax element for an input source width;
- decoding, from the bitstream, a syntax element for an input source height;
- decoding, from the bitstream, unpacking information for producing feature maps from a frame, wherein the unpacking information includes width information indicating a width of feature maps produced by an unpacking process, and
- height information indicating a height of feature maps produced by an unpacking process;
- producing, from the bitstream, a frame including feature maps;
- producing one or more tensors including the feature maps by performing an unpacking process on the produced frame, wherein the feature maps have the width and height indicated by the unpacking information;
- producing one or more reconstructed tensors by performing a reconstruction process on the one or more tensors including the feature maps, wherein the reconstruction process includes at least inverse normalization;
- and
- wherein one or more task results are to be produced by executing a part of Neural Network based on the one or more reconstructed tensor, the input source height, and the input source width.
6 more elements
A decoding apparatus comprising:
- a first decoding unit for decoding, from a bitstream, a syntax element for an input source width;
- a second decoding unit for decoding, from the bitstream, a syntax element for an input source height;
- a third decoding unit for decoding, from the bitstream, unpacking information for producing feature maps from a frame, wherein the unpacking information includes width information indicating a width of feature maps produced by an unpacking process, and
- height information indicating a height of feature maps produced by an unpacking process;
- a first producing unit for producing, from the bitstream, a frame including feature maps;
- a second producing unit for producing one or more tensors including the feature maps by performing an unpacking process on the produced frame, wherein the feature maps have the width and height indicated by the unpacking information;
- a third producing unit for producing one or more reconstructed tensors by performing a reconstruction process on the one or more tensors including the feature maps, wherein the reconstruction process includes at least inverse normalization;
- and
- wherein one or more task results are to be produced by executing a part of Neural Network based on the one or more reconstructed tensor, the input source_height, and the input source width.
6 more elements
A non-transitory computer-readable storage medium which stores a program for executing a method of decoding encoded data, the method comprising:
- decoding, from a bitstream, a syntax element for an input source width;
- decoding, from the bitstream, a syntax element for an input source height;
- decoding, from the bitstream, unpacking information for producing feature maps from a frame, wherein the unpacking information includes width information indicating a width of feature maps produced by an unpacking process, and
- height information indicating a height of feature maps produced by an unpacking process;
- producing, from the bitstream, a frame including feature maps;
- producing one or more tensors including the feature maps by performing an unpacking process on the produced frame, wherein the feature maps have the width and height indicated by the unpacking information;
- producing one or more reconstructed tensors by performing a reconstruction process on the one or more tensors including the feature maps, wherein the reconstruction process includes at least inverse normalization;
- and
- wherein one or more task results are to be produced by executing a part of Neural Network based on the one or more reconstructed tensor, the input source height, and the input source width.
6 more elements
Dependencies are inferred from the claim text.