Semi-Supervised Semantic Segmentation With High- and Low-Level Consistency

The ability to understand visual information from limited labeled data is an important aspect of machine learning. While image-level classification has been extensively studied in a semi-supervised setting, dense pixel-level with only drawn attention recently. In this work, we propose approach for semantic segmentation that learns pixel-wise annotated samples while exploiting additional annotation-free images. proposed relies on adversarial training feature matching loss learn unlabeled It uses two network branches link including self-training. dual-branch reduces both the low-level and high-level artifacts typical when few labels. attains significant improvement over existing methods, especially trained very samples. On several standard benchmarks-PASCAL VOC 2012, PASCAL-Context, Cityscapes-the achieves new state-of-the-art