CSCNet: Contextual semantic consistency network for trajectory prediction in crowded spaces

作者：

Highlights：

• We design a context-aware transfer to break the “Gap” within social and physical descriptions by aligning activity and scene semantics.

• We use a novel conditional context loss to train the whole network end-to-end to make predictions in line with social and physical rules.

• Combining the context-aware transfer and conditional context loss, CSCNet outperforms the existing models on ETH-UCY and SDD Datasets.

摘要

•We design a context-aware transfer to break the “Gap” within social and physical descriptions by aligning activity and scene semantics.•We use a novel conditional context loss to train the whole network end-to-end to make predictions in line with social and physical rules.•Combining the context-aware transfer and conditional context loss, CSCNet outperforms the existing models on ETH-UCY and SDD Datasets.

论文关键词：Trajectory prediction,The context-aware transfer,The conditional context loss

论文评审过程：Received 12 May 2021, Revised 20 January 2022, Accepted 22 January 2022, Available online 12 February 2022, Version of Record 16 February 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108552