Context-based image explanations for deep neural networks
作者：
Highlights：
• A model-agnostic method for generating context-based explanations using semantic segmentation for DNNs.
• Present visual and text-based explanations showing the relative importance of each feature for a prediction.
• Evaluate the explanation method using a user study to understand human perception on different explanations.
• Our explanation method visually out-performed existing gradient and occlusion based methods.
摘要
Highlights•A model-agnostic method for generating context-based explanations using semantic segmentation for DNNs.•Present visual and text-based explanations showing the relative importance of each feature for a prediction.•Evaluate the explanation method using a user study to understand human perception on different explanations.•Our explanation method visually out-performed existing gradient and occlusion based methods.
论文关键词：DNNs,Explainable AI,Contextual importance,Visual explanations
论文评审过程：Received 9 June 2021, Revised 13 August 2021, Accepted 10 September 2021, Available online 22 September 2021, Version of Record 30 September 2021.
论文官网地址：https://doi.org/10.1016/j.imavis.2021.104310
