Table 1.

Performance of different deep learning architectures

ArchitectureNo. of layersShort description
AlexNet8A landmark architecture for deep learning winning ILSVRC 2012 challenge (31).
NiN16Network in Network (NiN) is one of the first architectures harnessing innovative 1 × 1 convolutions (49) to provide more combinational power to the features of a convolutional layers (49).
VGG22An architecture that is deeper (i.e., has more layers of neurons) and obtains better performance than AlexNet by using effective 3 × 3 convolutional filters (26).
GoogLeNet32This architecture is designed to be computationally efficient (using 12 times fewer parameters than AlexNet) while offering high accuracy (50).
ResNet18, 34, 50, 101, 152The winning architecture of the 2016 ImageNet competition (25). The number of layers for the ResNet architecture can be different. In this work, we try 18, 34, 50, 101, and 152 layers.