Progressive Token Length Scaling in Transformer Encoders for Efficient Universal Segmentation

arXiv:2404.14657v1 Announce Type: new
Abstract: A powerful architecture for universal segmentation relies on transformers that encode multi-scale image features and decode object queries into mask predictions. With efficiency being a high priority for scaling such models, we observed that the state-of-the-art method Mask2Former uses ~50% of its compute only on the transformer encoder. This is due to the retention of a full-length token-level representation of all backbone feature scales at each encoder layer. With this observation, we propose a strategy …

abstract architecture art arxiv compute cs.cv decode efficiency encode features image mask2former object predictions queries scale scaling segmentation state token transformer transformers type universal