Name

dnn_topstim - For maximizing activation of a specific layer and channel in DNN network, selecting the topK stimuli from a stimulus set. Meanwhile, save their activation.

Synopsis

dnn_topstim [-h] -net Net -top TopNumber -stim Stimulus -layer Layer
            [Layer ...] -chn Channel [Channel ...] [-cuda] -out OutputDir

Arguments

Required Arguments

Argument	Discription
net	Name of DNN Model, which should be placed in $DNNBRAIN_DATA/models with format *.pth.
top	Number of top stimulus. For example, assign top=5, and top 5 image for each <layer,channel> pair will be selected.
stim	a .stim.csv file which contains stimulus information
layer	names of the layers used to specify where activation is extracted fromFor example, ‘conv1’ represents the first convolution layer, and ‘fc1’ represents the first full connection layer.
chn	Channel numbers used to specify where activation is extracted from
out	Output directory to save .stim.csv for top stimulus, and associated .act.hd5 file.

Optional Arguments

Argument	Discription
cuda	Use GPU or not

Outputs

several stim.csv files that contain information of top stimuli of each channel a .act.h5 file that contains raw activation of the top stimuli

Examples

Select top3 stimuli for 1st, 2nd, and 3rd channels of conv5 layer of AlexNet respectively.

dnn_topstim -net AlexNet -top 3 -stim examples.stim.csv -layer conv5 -chn 1 2 3 -out .