Computer Vision News - January 2022

22 Deep Learning Tool To re-create vgg16, we needed to add the definition of a Fully Connected layer in the file pycore/ tikzeng.py. This file writes up a drawing for each layer in latex code. If layers for your chosen architecture are missing here, you will need to create them yourself. This can be quite time consuming, but I believe the result is worth it. You can judge by yourselves on the diagram below. # Fully Connected def to_FullyConnected ( name, s_filer=” “, n_filer=” “, offset=”(0,0,0)”, to=”(0,0,0)”, width= 1.5 , height= 3 , depth= 25 , opacity= 0.8 , caption=” “): return r””” \pic[shift={“””+ offset +”””}] at “””+ to +””” {Box={ name=””” + name +”””, caption=””” +caption + “””, xlabel={{ “””+ ‘”’+str(n_filer) +’”, “dummy”’+ “”” }}, zlabel=”””+ str(s_filer) +”””, fill=\FcColor, opacity=”””+ str(opacity) +”””, height=”””+ str(height) +”””, width=”””+ str(width) +”””, depth=”””+ str(depth) +””” } }; “”” Finally, run the script with: bash .. /tikzmake.sh my_arch Once you run this, a pdf file containing the image of your network will be automatically saved. If you also need to store the .tex version, for use on Overleaf or another latex code, please make sure you go to .. /tikxmake.sh and comment the line “rm *.tex”. 224224 64 conv1 112 112 128 conv2 56 56 256 conv3 56 256 28 28 512 conv4 28 512 14 14 512 conv5 14 512 4096 fc6 4096 fc7 1000 fc8+ softmax 1000 Diagram of vgg16 architecture created using PlotNeuralNet