Object Segmentation using Pixel-wise Adversarial Loss

TL;DR

This paper introduces a novel pixel-wise adversarial loss combined with stochastic weight averaging to enhance deep learning-based object segmentation, achieving state-of-the-art results.

Contribution

It proposes a new pixel-wise adversarial training method and employs stochastic weight averaging to improve segmentation accuracy over existing models.

Findings

Significant performance improvements over baseline models.

State-of-the-art segmentation results achieved.

Pixel-wise adversarial loss enhances detail in segmentation maps.

Abstract

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.