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.
@article{arxiv.1909.10341,
title = {Object Segmentation using Pixel-wise Adversarial Loss},
author = {Ricard Durall and Franz-Josef Pfreundt and Ullrich Köthe and Janis Keuper},
journal= {arXiv preprint arXiv:1909.10341},
year = {2019}
}