Targeted data augmentation using neural style transfer

A method for training a deep neural network (DNN) to perform a specified task with respect to images having a target image style, comprising:

• (a) obtaining the DNN to be trained;
• (b) obtaining an original training dataset, which includes a plurality of original training images and a training label for each original training image, the original training images having image styles different from the target image style;
• (c) obtaining an image having the target image style;
• (d) using the image having the target image style as a style target image, training a style transformer network to perform a style transformation that transforms any photorealistic input image into a transformed image that has contents of the input image, maintains photorealistic quality of the input image, and has an image style that matches the image style of the style target image, wherein the style transformer network is distinct from the DNN to be trained;
• (e) using the trained style transformer network to transform each of the original training images into a corresponding transformed training image;
• (f) labeling each transformed training image with the training label of the corresponding original training image to form an augmented training dataset;
• and
• (g) using the augmented training dataset to train the DNN to perform the specified task.