Efficient densely connected convolutional neural networks

作者：

Highlights：

• Proposed two efficient densely connected ConvNets, DesneDsc and Dense2Net.

• DesneDsc and Dense2Net are more efficient and have higher accuracy than DenseNet.

• Dense2Net are state-of-the-art on ImageNet in manual CNNs with <10 M parameters.

• DenseDsc and Dense2Net are very flexible and can be used in different applications.

摘要

•Proposed two efficient densely connected ConvNets, DesneDsc and Dense2Net.•DesneDsc and Dense2Net are more efficient and have higher accuracy than DenseNet.•Dense2Net are state-of-the-art on ImageNet in manual CNNs with <10 M parameters.•DenseDsc and Dense2Net are very flexible and can be used in different applications.

论文关键词：Convolutional neural networks,Classification,Parameter efficiency,Densely connected

论文评审过程：Received 2 September 2019, Revised 30 July 2020, Accepted 19 August 2020, Available online 20 August 2020, Version of Record 23 August 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107610