TY  - CHAP
U1  - Konferenzveröffentlichung
A1  - Durall Lopez, Ricard
A1  - Pfreundt, Franz-Josef
A1  - Keuper, Janis
T1  - Object Segmentation Using Pixel-Wise Adversarial Loss
T2  - DAGM GCPR 2019: Pattern Recognition
N2  - 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.
Y1  - 2019
SN  - 1611-3349 (Online)
SS  - 1611-3349 (Online)
SN  - 0302-9743 (Print)
SS  - 0302-9743 (Print)
SN  - 978-3-030-33676-9 (Online)
SB  - 978-3-030-33676-9 (Online)
SN  - 978-3-030-33675-2 (Print)
SB  - 978-3-030-33675-2 (Print)
U6  - https://doi.org/10.1007/978-3-030-33676-9_21
DO  - https://doi.org/10.1007/978-3-030-33676-9_21
VL  - LNCS 11824
SP  - 303
EP  - 316
PB  - Springer
CY  - Cham
ER  -