Name¶
dnn_rf_sw - Visualize receptive field using slide-window(sw) occulding method for interested channels
Synopsis¶
dnn_rf_sw -net Net -layer Layer -chn Channel -stim Stimulus -wsize WindowSize -stride Stride -metric Metric -out Output
Arguments¶
Required Arguments¶
Argument |
Discription |
|---|---|
net |
Name of a neural network |
stim |
Path of a .stim.csv file which contains stimulus information |
layer |
Name of the target layer.Only support one layer each time.E.g., ‘conv1’ represents the first convolution layer, and ‘fc1’ represents the first full connection layer. |
chn |
Index of target channel.Only support one channel each time.Channel index starts from 1. |
wsize |
Windows size of occluder.Please enter two integers which define window’s length and width. |
stride |
Stride of occluder window.Please enter two integers which define stride in x_axis and y_axis. |
metric |
Metric to extract the unit’s activation.Only support max/mean/L1/L2. |
out |
Output path to save the occluder image |
Outputs¶
A series of occluder images based on your stimilus and interested net information
Examples¶
dnn_rf_sw -net AlexNet -layer conv5 -chn 122 -stim ./flowers.stim.csv -wsize 11 11 -stride 2 2 -metric max -out ./image/rf_sw/
The original image used in this doc is displayed as below:
The receptive field occulding image is displayed as below:
