Deep ancient Roman Republican coin classification via feature fusion and attention

作者：

Highlights：

• Motif recognition on ancient Roman coins with 98.5% accuracy on 100 classes.

• Largest ancient coin dataset presented so far.

• Compact Linear Pooling of two models’ feature output for better performance.

• Generalization ability of method is demonstrated on 128 unseen classes.

摘要

•Motif recognition on ancient Roman coins with 98.5% accuracy on 100 classes.•Largest ancient coin dataset presented so far.•Compact Linear Pooling of two models’ feature output for better performance.•Generalization ability of method is demonstrated on 128 unseen classes.

论文关键词：Coins dataset,Compact bilinear pooling,Convolutional networks,Visual attention,Residual blocks,Deep learning in art’s history,Roman Republican coins

论文评审过程：Received 2 September 2019, Revised 6 December 2020, Accepted 31 January 2021, Available online 3 February 2021, Version of Record 12 February 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107871