Semantic segmentation of sparse multi-dimensional tensors

One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:

• extracting, from a set of n-dimensional images, a first set of n-dimensional patches having a first patch size;
• during a first training stage, training a machine learning model using the first set of n-dimensional patches and a first set of losses associated with the first set of n-dimensional patches;
• and
• upon determining that the first set of losses meets one or more thresholds corresponding to a performance of the machine learning model:
• determining, based on the performance of the machine learning model, a second patch size that is (i) associated with a second set of n-dimensional patches extracted from the set of n-dimensional images and (ii) smaller than the first patch size;
• and
• during a second training stage following the first training stage, training the machine learning model using the second set of n-dimensional patches and a second set of losses associated with the second set of n-dimensional patches.

A system, comprising:

• one or more memories that store instructions, and
• one or more processors that are coupled to the one or more memories and, when executing the instructions, are configured to perform the steps of:
• extracting, from a set of n-dimensional images, a first set of n-dimensional patches having a first patch size;
• training a machine learning model using a first set of losses associated with the first set of n-dimensional patches;
• and
• upon determining that the first set of losses meets one or more thresholds corresponding to a performance of the machine learning model:
• determining, based on the performance of the machine learning model, a second patch size that is (i) associated with a second set of n-dimensional patches extracted from the set of n-dimensional images and (ii) smaller than the first patch size;
• and
• training the machine learning model using the second set of n-dimensional patches and a second set of losses associated with the second set of n-dimensional patches.

A computer-implemented method for performing semantic segmentation, the method comprising:

• extracting a first set of n-dimensional patches from a n-dimensional image;
• executing a trained machine learning model to convert the first set of n-dimensional patches into a set of n-dimensional semantic segmentations, wherein the trained machine learning model is generated by:
• training a machine learning model using a second set of n-dimensional patches having a first patch size and a first set of losses;
• upon determining that the first set of losses meets one or more thresholds corresponding to a performance of the machine learning model:
• determining, based on the performance of the machine learning model, a second patch size that is (i) associated with a third set of n-dimensional patches and (ii) smaller than the first patch size;
• and
• training the machine learning model using the third set of n-dimensional patches and a second set of losses;
• and
• generating a semantic segmentation of the n-dimensional image based on the set of n-dimensional semantic segmentations.