Adversarial Learning for Semi-Supervised Semantic Segmentation

We propose a method 1 for semi-supervised semantic segmentation using the adversarial network. While most existing discriminators are trained to classify input images as real or fake on the image level, we design a discriminator in a fully convolutional manner to differentiate the predicted probability maps from the ground truth segmentation distribution with the consideration of the spatial resolution. We show that the proposed discriminator can be used to improve the performance on semantic segmentation by coupling the adversarial loss with the standard cross entropy loss on the segmentation network. In addition, the fully convolutional discriminator enables the semi-supervised learning through discovering the trustworthy regions in prediction results of unlabeled images, providing additional supervisory signals. In contrast to existing methods that utilize weakly-labeled images, our method leverages unlabeled images without any annotation to enhance the segmentation model. Experimental results on both the PASCAL VOC 2012 dataset and the Cityscapes dataset demonstrate the effectiveness of our algorithm.