SEMANTIC SEGMENTATION GROUND TRUTH CORRECTION WITH SPATIAL TRANSFORMER NETWORKS

An apparatus accesses label data and training images corresponding to a geographic area; and provides the label data and training images to a training model. The training model comprises of at least a predictor model and an alignment model. The predictor model is configured to receive an image and provide a prediction corresponding to the image. The alignment model is configured to generate a transformed prediction based on aligning the label data and the prediction. The apparatus executes a loss engine to iteratively receive the label data and the transformed prediction, evaluate a loss function based on the label data and the transformed prediction, and cause weights of the predictor model and the alignment model to be updated based on the evaluated loss function to cause the predictor and alignment models to be trained.