Enhanced training method and apparatus for image recognition model

A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:

• randomly selecting a predetermined quantity of samples from a first sample set as seed samples, wherein a seed sample of the seed samples corresponds to a seed image and an annotation result for the seed image;
• obtaining, for each of the seed samples, at least one extended sample based on an extension operation for the seed sample, wherein the extension operation comprises:
• separately merging, on at least one image channel, pixels of the seed image of the seed sample with disturbance values to obtain a disturbance image, wherein the disturbance values are arranged in a predetermined distribution with a reference pixel as a center, and the predetermined distribution is related to a distance from each of the pixels to the reference pixel;
• and
• constructing an extended sample based on the disturbance image and the annotation result for the seed image, wherein the reference pixel is determined based on a predetermined rule;
• and
• performing an enhanced training in an image recognition model using the at least one extended sample as input.

A computer-implemented system, comprising:

• one or more computers;
• and
• one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform randomly selecting a predetermined quantity of samples from a first sample set as seed samples, wherein a seed sample of the seed samples corresponds to a seed image and an annotation result for the seed image;
• obtaining, for each of the seed samples, at least one extended sample based on an extension operation for the seed sample, wherein the extension operation comprises:
• separately merging, on at least one image channel, pixels of the seed image of the seed sample with disturbance values to obtain a disturbance image, wherein the disturbance values are arranged in a predetermined distribution with a reference pixel as a center, and the predetermined distribution is related to a distance from each of the pixels to the reference pixel;
• and
• constructing an extended sample based on the disturbance image and the annotation result for the seed image, wherein the reference pixel is determined based on a predetermined rule;
• and
• performing an enhanced training in an image recognition model using the at least one extended sample as input.