Method, computer program and device for training a dynamic-architecture convolutional neural network for semantic image segmentation

Abstract

The invention relates to a method for training a dynamic-architecture convolutional neural network for semantic image segmentation, including auto-annotation of images, to constitute a first image database of auto-annotated images. The method includes applying at least one transformation to at least one source image, and adding, to each transformed image thus obtained, a first tag corresponding to the transformation. The method also includes at least one iteration of a phase including training the neural network with the first image database, and a second image database annotated with a second tag relating to the content of the image. The invention also relates to a computer program and a device implementing the method, and a deep-learning dynamic-architecture convolutional neural network obtained by the method.