Patent attributes
A method, apparatus and computer program product are provided for augmenting a training data set. In a method, a first type of road sign is identified within an image, the orientation and scale of the first type of road sign within the image are estimated and stylistic content associated with the first type of road sign within the image is identified. The method transforms an image of a second, less common type of road sign based upon the orientation, scale and stylistic content of the first type of road sign. The method also creates a synthetic image in which the first type of road sign is replaced by a transformed representation of the second type of road sign and pixels are filled in the synthetic image about the second type of road sign that were previously occluded by the first type of road sign.
