Layer to Module

DNNBrain allows users to conveniently approach to the module inside a built-in DNN by a commonly used layer name. For each built-in DNN, its available layer names and the mapping between layer names and modules are shown as follow. And we take the AlexNet as example to interpret mechanism of the mapping.

AlexNet

Generally, we can approach to a module by its location in the DNN framework. For example, according to AlexNet framework (Fig. 1, left), we can locate the first convolutional layer by its location (features, 0). However, DNNBrain users can easily do it by the layer name ‘conv1’ according to the mapping (Fig. 1, right).

AlexNet_framework AlexNet_mapping

Figure 1

VggFace

VggFace_framework VggFace_mapping

Figure 2

Vgg11

Vgg11_framework Vgg11_mapping

Figure 3

Vgg19_bn

Vgg19_bn_framework Vgg19_bn_mapping

Figure 4

GoogLeNet

GoogLeNet_mapping

Figure 5

Resnet152

resnet152_mapping

Figure 6