A survey on deep learning methods for scene flow estimation

作者：

Highlights：

• We review some of the most important recent works on scene flow estimation with standardized description, and categorize them into three categories.

• We give the formal definition for scene flow estimation, including optical flow and depth estimation.

• We discuss the most prevalent basic architecture for scene flow estimation and summarize them into four main steps.

• We present the performance comparison among different methods.

摘要

•We review some of the most important recent works on scene flow estimation with standardized description, and categorize them into three categories.•We give the formal definition for scene flow estimation, including optical flow and depth estimation.•We discuss the most prevalent basic architecture for scene flow estimation and summarize them into four main steps.•We present the performance comparison among different methods.

论文关键词：Scene flow,Optical flow,Depth estimation,Deep learning

论文评审过程：Received 30 September 2019, Revised 30 March 2020, Accepted 12 April 2020, Available online 16 May 2020, Version of Record 16 May 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107378