Vision transformers for dense prediction: A survey

作者：

Highlights：

• We provide a comprehensive review of state-of-the-art transformer methods.

• We focus on the transformer-based methods in the area of dense prediction tasks.

• We propose a model taxonomy according to architectures and optimizations.

• We conduct a systematic horizontal comparison of multitudinous methods.

摘要

•We provide a comprehensive review of state-of-the-art transformer methods.•We focus on the transformer-based methods in the area of dense prediction tasks.•We propose a model taxonomy according to architectures and optimizations.•We conduct a systematic horizontal comparison of multitudinous methods.

论文关键词：Deep learning,Transformer,Dense prediction,Computer vision

论文评审过程：Received 10 May 2022, Revised 19 July 2022, Accepted 22 July 2022, Available online 28 July 2022, Version of Record 6 August 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109552