Distance transform regression for spatially-aware deep semantic segmentation

Nicolas Audebert (1, 2) , Alexandre Boulch (1) , Bertrand Le Saux (1) , Sébastien Lefèvre (2)

1 DTIS, ONERA, Université Paris Saclay (COmUE) [Palaiseau]
2 OBELIX - Environment observation with complex imagery

Understanding visual scenes relies more and more on dense pixel-wise classification obtained via deep fully convolutional neural networks. However, due to the nature of the networks, predictions often suffer from blurry boundaries and ill-segmented shapes, fueling the need for post-processing. This work introduces a new semantic segmentation regularization based on the regression of a distance transform. After computing the distance transform on the label masks, we train a FCN in a multi-task setting in both discrete and continuous spaces by learning jointly classification and distance regression. This requires almost no modification of the network structure and adds a very low overhead to the training process. Learning to approximate the distance transform back-propagates spatial cues that implicitly regularizes the segmentation. We validate this technique with several architectures on various datasets, and we show significant improvements compared to competitive baselines.

Domaines: Réseau de neurones [cs.NE]; Vision par ordinateur et reconnaissance de formes [cs.CV]

Nicolas Audebert, Alexandre Boulch, Bertrand Le Saux, Sébastien Lefèvre. Distance transform regression for spatially-aware deep semantic segmentation. Computer Vision and Image Understanding, 2019, 189, pp.102809. ⟨10.1016/j.cviu.2019.102809⟩. ⟨hal-02277621⟩