Open AccessA New Lightweight DenseNet Based on Mix-Structure Convolution
Author(s) -
Yi He
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/790/1/012113
Subject(s) - flops , convolution (computer science) , computer science , redundancy (engineering) , algorithm , parallel computing , artificial intelligence , artificial neural network , operating system
In this paper, we propose a new and efficient lightweight DenseNet which optimizes the parameter redundancy and high FLOPs in the DenseNet model. According to the distribution of the weight value, the element of the Lightweight Mix-Structure Convolution (LMSC) is realized in the model, which reduces the calculation and parameter required for model construction, and ensures that the accuracy does not decline significantly. The experimental results show that compared with the DenseNet-40-P model the model only uses 45.5% of the parameters and 54.3% of the FLOPs, and the accuracy is only reduced by less than 0.4%.