REPRODUCING AUGMENTATION OF IMAGE DATA Russian patent published in 2020 - IPC G06N3/08 G06T5/00 

FIELD: physics.

SUBSTANCE: invention relates to a method and a system for random augmentation of data for use in training machine learning models. Method comprises steps of: obtaining one or more first images associated with a training sample of images for training machine learning model; providing one or more first images as a first input for a first plurality of layers of computing units, wherein first plurality of layers uses image filters; providing first result of first plurality of layers of computing units as second input for second layer of computing units, wherein second layer uses random sets of parameters for calculations; obtaining distortion parameters from the second layer of computing units; creating one or more second images based on one or more first images and distortion parameters; obtaining as third output one or more second images and adding one or more second images to a training sample of images for training machine learning model.

EFFECT: high efficiency of generating training data samples for machine learning model without limiting the number of augmentations of the original image.

20 cl, 11 dwg