Semi-Supervised Semantic Segmentation with Pixel-Level Contrastive Learning from a Class-wise Memory Bank

This work presents a novel approach for semi-supervised semantic\nsegmentation. The key element of this approach is our contrastive learning\nmodule that enforces the segmentation network to yield similar pixel-level\nfeature representations for same-class samples across the whole dataset. To\nachieve this, we maintain a memory bank continuously updated with relevant and\nhigh-quality feature vectors from labeled data. In an end-to-end training, the\nfeatures from both labeled and unlabeled data are optimized to be similar to\nsame-class samples from the memory bank. Our approach outperforms the current\nstate-of-the-art for semi-supervised semantic segmentation and semi-supervised\ndomain adaptation on well-known public benchmarks, with larger improvements on\nthe most challenging scenarios, i.e., less available labeled data.\nhttps://github.com/Shathe/SemiSeg-Contrastive\n