Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Object Segmentation Using Pixel-Wise Adversarial Loss

: Durall, Ricard; Pfreundt, Franz-Josef; Köthe, Ullrich; Keuper, Janis


Fink, G.A. ; Deutsche Arbeitsgemeinschaft für Mustererkennung -DAGM-:
Pattern Recognition : 41st DAGM German Conference, DAGM GCPR 2019, Dortmund, Germany, September 10-13, 2019 : proceedings
Cham: Springer, 2019 (Lecture Notes in Computer Science 11824)
ISBN: 978-3-030-33676-9
ISBN: 3-030-33675-1
ISBN: 978-3-030-33676-9
German Conference on Pattern Recognition (GCPR) <41, 2019, Dortmund>
Conference Paper
Fraunhofer ITWM ()

Recent deep learning based approaches have shown remarkable success on object segmentation tasks. However, there is still room for further improvement. Inspired by generative adversarial networks, we present a generic end-to-end adversarial approach, which can be combined with a wide range of existing semantic segmentation networks to improve their segmentation performance. The key element of our method is to replace the commonly used binary adversarial loss with a high resolution pixel-wise loss. In addition, we train our generator employing stochastic weight averaging fashion, which further enhances the predicted output label maps leading to state-of-the-art results. We show, that this combination of pixel-wise adversarial training and weight averaging leads to significant and consistent gains in segmentation performance, compared to the baseline models.