Explaining the semantics capturing capability of scene graph generation models

作者：

Highlights：

• Propose metrics for explaining the semantics capturing capability of scene graph generation DNNs.

• Support nine semantic properties in OWL2 for specifying property axioms.

• Help revealing biases of DNN models in handling relations with different semantic properties.

• Verify that DNNs can capture semantic properties of relations even with incomplete annotations.

摘要

•Propose metrics for explaining the semantics capturing capability of scene graph generation DNNs.•Support nine semantic properties in OWL2 for specifying property axioms.•Help revealing biases of DNN models in handling relations with different semantic properties.•Verify that DNNs can capture semantic properties of relations even with incomplete annotations.

论文关键词：Explanation,Metrics,Semantic property,Scene graph generation,Deep neural network

论文评审过程：Received 5 August 2019, Revised 11 April 2020, Accepted 5 May 2020, Available online 30 May 2020, Version of Record 1 November 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107427