Superpixel methods for convolutional neural networks

A system comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations for more efficiently performing a computation of a convolutional neural network layer that is configured to receive an X by Y by Z input tensor and process the received input tensor using one or more convolutional neural network layer weight matrices to generate a U by V by W output tensor, the operations comprising:

• receiving an X by Y by Z input tensor, the input tensor comprising XY first inputs of depth Z wherein X and Y represent spatial extent and Z represents depth extent;
• generating an X′ by Y′ by Z′ modified input tensor, wherein X′ is smaller than or equal to X, Y′ is smaller than or equal to Y, and Z′ is larger than or equal to Z, comprising:
• grouping together two or more of the first inputs;
• and
• generating X′Y′ super inputs by converting spatial extent of the grouped first inputs to depth extent;
• obtaining one or more modified convolutional neural network layer weight matrices, wherein (i) the modified weight matrices operate on the X′ by Y′ by Z′ modified input tensor to generate a U′ by V′ by W′ modified output tensor, and (ii) the U′ by V′ by W′ output tensor comprises a transformed U by V by W output tensor, wherein U′ is smaller than or equal to U, V′ is smaller than or equal to V, and W′ is larger than or equal to W;
• and
• performing a computation of a modified convolutional neural network layer having the modified weight matrices on the modified input to generate the U′ by V′ by W′ output tensor, wherein the U′ by V′ by W′ output tensor comprises the U by V by W output that corresponds to an output generated by the convolutional neural network layer when processing the X by Y by Z input tensor.