Method for image processing by means of a neural network and method for training a neural network

A computer-implemented method for training a neural network, comprising:

• reading an image that is divided into a plurality of image regions;
• grouping at least two image regions of the plurality of image regions into at least two groups based on one or more properties of the plurality of image regions, wherein each of the plurality of image regions is assigned to exactly one group or to no group and image regions with the same or similar properties are assigned to the same group;
• computing partial results for the image regions of the at least two groups by:
• applying at least two kernels in a layer of the neural network to the image regions of the at least two groups, wherein at most one kernel is applied to the image regions of each group;
• and
• generating a result matrix of a layer of the neural network based on the partial results for the image regions of the at least two groups, wherein a data volume of the result matrix is less than a data volume that would be produced by applying each of the at least two kernels to each image region of the plurality of image regions.

A device for processing an image in a neural network, comprising:

• at least one memory, wherein at least one image is stored in at the least one memory;
• and
• at least one processor, wherein the at least one processor is configured to:
• divide the image into a plurality of image regions;
• group at least two image regions of the plurality of image regions into at least two groups based on one or more properties of the plurality of image regions, wherein each of the plurality of image regions is assigned to exactly one group or to no group and image regions with the same or similar properties are assigned to the same group;
• compute partial results for the image regions of the at least two groups by being further configured to:
• apply at least two kernels in a layer of the neural network to the image regions of the at least two groups, wherein at most one kernel is applied to the image regions of each group;
• and
• generate a result matrix of a layer of the neural network based on the partial results for the image regions of the at least two groups, wherein a data volume of the result matrix is less than a data volume that would be produced by applying each of the at least two kernels to each image region of the plurality of image regions.