Texture Compression Using a Neural Network

A computer-implemented method comprising:

• receiving a pixel region of a texture map;
• generating, using a compression parameter neural network,
• one or more compression parameters for a compressor;
• compressing, using the compressor, the pixel region to a compressed representation, comprising compressing the pixel region in accordance with the one or more compression parameters;
• and
• storing the compressed representation of the pixel region of the texture map to one or more memories accessible by a graphics processing unit.

A computer-implemented method for training a compression parameter neural network for use in compression of a texture map, comprising:

• receiving a plurality of training pixel regions and a plurality of respective gold standard compression parameter sets;
• inputting a vector representation of one of said plurality of training pixel regions to a compression parameter neural network, wherein the compression parameter neural network comprises a plurality of neural network layers and a plurality of neural network weights;
• in response to the inputting, generating a compression parameter set for said pixel region using the compression parameter neural network;
• and
• training the compression parameter neural network, comprising updating one or more neural network weights of the plurality of neural network weights to minimise a loss measure based at least in part on a comparison between the generated compression parameter set and the corresponding gold standard compression parameter set.

A non-transitory computer readable storage medium storing computer program code that, when executed by a processing device, cause the processing device to perform operations comprising:

• receiving a texture map;
• segmenting the texture map into a plurality of pixel regions;
• for each of the plurality of pixel regions:
• inputting a vector representation of the pixel region to a compression parameter neural network, wherein the compression parameter neural network is configured to:
• process the vector representation of the pixel region through one or more layers of the compression parameter neural network;
• and
• generate an output indicating a compression parameter set for compressing the pixel region;
• inputting the pixel region and the compression parameter set to a compressor, wherein the compressor is configured to compress the pixel region to a compressed representation of the pixel region in accordance with the compression parameter set;
• storing a compressed representation of the texture map to a memory accessible by a graphics processing unit, wherein storing the compressed representation of the texture map comprises storing the compressed representations of each of the pixel regions to said memory;
• and
• selectively decompressing portions of the compressed representation of the texture map using the graphics processing unit.

A non-transitory computer readable storage medium storing computer program code that, when executed by a processing device, cause the processing device to perform operations comprising:

• receiving a pixel region of a texture map;
• generating, using a compression parameter neural network,
• one or more compression parameters for a compressor;
• compressing, using the compressor, the pixel region to a compressed representation, comprising compressing the pixel region in accordance with the one or more compression parameters;
• and
• storing the compressed representation of the pixel region of the texture map to one or more memories accessible by a graphics processing unit.

An apparatus comprising:

• at least one processor;
• and
• at least one memory including computer program code which, when executed by the at least one processor, causes the apparatus to:
• receive a plurality of training pixel regions and a plurality of respective gold standard compression parameter sets;
• input a vector representation of one of said plurality of training pixel regions to a compression parameter neural network, wherein the compression parameter neural network comprises a plurality of neural network layers and a plurality of neural network weights;
• in response to the inputting of the vector representation, generate a compression parameter set for said pixel region using the compression parameter neural network;
• and
• train the compression parameter neural network, wherein training the compression parameter neural network comprises updating one or more neural network weights of the plurality of neural network weights to minimise a loss measure based at least in part on a comparison between the generated compression parameter set and the corresponding gold standard compression parameter set.