Fusion global and local deep representations with neural attention for aesthetic quality assessment

作者：

Highlights：

• A global–local fusion network is proposed for image aesthetic assessment, which can support learning from arbitrary image sizes.

• We explore the possibility to extract local fine-grained features based on the top-down attention of the neural network.

• We combine the strength of classification and regression by a multi-task learning framework, which make the network converge faster.

摘要

•A global–local fusion network is proposed for image aesthetic assessment, which can support learning from arbitrary image sizes.•We explore the possibility to extract local fine-grained features based on the top-down attention of the neural network.•We combine the strength of classification and regression by a multi-task learning framework, which make the network converge faster.

论文关键词：Image quality assessment,Image aesthetics analysis,Deep neural network

论文评审过程：Received 29 May 2018, Revised 31 May 2019, Accepted 31 May 2019, Available online 13 June 2019, Version of Record 17 June 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.05.021