Scaling up Multi-domain Semantic Segmentation with Sentence Embeddings

arXiv:2202.02002v2 Announce Type: replace
Abstract: We propose an approach to semantic segmentation that achieves state-of-the-art supervised performance when applied in a zero-shot setting. It thus achieves results equivalent to those of the supervised methods, on each of the major semantic segmentation datasets, without training on those datasets. This is achieved by replacing each class label with a vector-valued embedding of a short paragraph that describes the class. The generality and simplicity of this approach enables merging multiple datasets from different …

abstract art arxiv cs.cv datasets domain embeddings major performance results scaling scaling up segmentation semantic state training type zero-shot