Semantic Segmentation via Multi-task, Multi-domain Learning

Damien Fourure (1, 2) , Rémi Emonet (2) , Elisa Fromont (2) , Damien Muselet (2) , Alain Trémeau (2) , Christian Wolf (1)

1 imagine - Extraction de Caractéristiques et Identification
2 LHC - Laboratoire Hubert Curien

We present an approach that leverages multiple datasets possibly annotated using different classes to improve the semantic segmentation accuracy on each individual dataset. We propose a new selective loss function that can be integrated into deep networks to exploit training data coming from multiple datasets with possibly different tasks (e.g., different label-sets). We show how the gradient-reversal approach for domain adaptation can be used in this setup. Thorought experiments on semantic segmentation applications show the relevance of our approach.

Mots clés: Deep learning; Convolutional neural networks; Semantic segmentation; Domain adaptation; Multi-task learning

Domaines: Réseau de neurones [cs.NE]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]

Damien Fourure, Rémi Emonet, Elisa Fromont, Damien Muselet, Alain Trémeau, et al.. Semantic Segmentation via Multi-task, Multi-domain Learning. S+SSPR 2016 The joint IAPR International Workshops on Structural and Syntactic Pattern Recognition (SSPR 2016) and Statistical Techniques in Pattern Recognition (SPR 2016) , Nov 2016, Merida, Mexico. ⟨hal-01376998⟩