Training dataset augmentation method and system for training deep learning network

A training dataset augmentation system comprising:

• one or more processors comprising:
• a configurator configured to configure a training dataset with image data from which a specific area is detected, and a label, wherein the specific area is detected by using a detector that is trained to detect a predefined area from an input image data;
• and
• a generator configured to add a new training dataset by transforming the detected area in the configured training dataset,
• wherein, for the adding, the one or more processors are further configured to extract coordinates of the detected area from the configured training dataset, generate a random number, and update the coordinates of the detected area based on the generated random number,
• wherein, for the updating, the one or more processors are further configured to update the coordinates for a top-left vertex of the detected area and a bottom-right vertex of the detected area based on different random numbers, and
• wherein the one or more processors are further configured to refine the updated coordinate in response to leftmost coordinates being determined to be larger than rightmost coordinates, or lowermost coordinates being determined to be larger than uppermost coordinates.

A deep learning model training method comprising:

• adding a new training dataset by transforming a detected area in a training dataset which is configured with image data from which a specific area is detected, and a label, wherein the specific area is detected by using a detector that is trained to detect a predefined area from an input image data;
• and
• training a deep learning model which performs inference from the image data from which the specific area is detected, by using the configured training dataset and the added training dataset,
• wherein the adding further comprises extracting coordinates of the detected area from the configured training dataset, generating a random number, and updating the coordinates of the detected area based on the generated random number, and
• wherein the updating comprises updating the coordinates for a top-left vertex of the detected area and a bottom-right vertex of the detected area based on different random numbers;
• and
• refining the updated coordinate in response to leftmost coordinates being determined to be larger than rightmost coordinates, or lowermost coordinates being determined to be larger than uppermost coordinates.