METHOD FOR GENERATING AN AUGMENTED DATA SET

A method for artificially generating semantically segmented layouts includes: defining a semantic code (204) comprising a vector of values, each corresponding to a respective proportion of a class in the semantically segmented layout; combining a noise vector (202) and the semantic code (204) into a feature map (Πn) with a first resolution; applying a softmax regression to produce normalized map (p); multiplying the values of each location of the normalized map (p) within a channel by a corresponding value for a class for the semantic code to produce a weighted feature map (ω); normalizing channel values at each location of the weighted feature map (ω) so that the value at each location in a channel represents a proportion of the total probability of channel values at the location to produce a candidate segmented layout (m); fusing and upscaling the candidate segmented layout (m) with the feature map (Πn) to provide a higher resolution feature map (Πn+1); and repeating the process to produce a higher resolution candidate segmented layout (m).