harisiqbal88 / plotneuralnet

Latex code for making neural networks diagrams

PlotNeuralNet

Latex code for drawing neural networks for reports and presentation. Have a look into examples to see how they are made. Additionally, lets consolidate any improvements that you make and fix any bugs to help more people with this code.

Examples

Following are some network representations:

FCN-8 (view on Overleaf)

FCN-32 (view on Overleaf)

Holistically-Nested Edge Detection (view on Overleaf)

Getting Started

1. Install the following packages on Ubuntu.

   • Ubuntu 16.04

     sudo apt-get install texlive-latex-extra
     

   • Ubuntu 18.04.2 Base on this website, please install the following packages.

     sudo apt-get install texlive-latex-base
     sudo apt-get install texlive-fonts-recommended
     sudo apt-get install texlive-fonts-extra
     sudo apt-get install texlive-latex-extra
     

   • Windows

   1. Download and install MikTeX.
   2. Download and install bash runner on Windows, recommends Git bash or Cygwin(https://www.cygwin.com/)

2. Execute the example as followed.

   cd pyexamples/
   bash ../tikzmake.sh test_simple
   

TODO

• Python interface
• Add easy legend functionality
• Add more layer shapes like TruncatedPyramid, 2DSheet etc
• Add examples for RNN and likes.

Latex usage

See examples directory for usage.

Python usage

First, create a new directory and a new Python file:

$ mkdir my_project
$ cd my_project
vim my_arch.py

Add the following code to your new file:

import sys
sys.path.append('../')
from pycore.tikzeng import *

# defined your arch
arch = [
    to_head( '..' ),
    to_cor(),
    to_begin(),
    to_Conv("conv1", 512, 64, offset="(0,0,0)", to="(0,0,0)", height=64, depth=64, width=2 ),
    to_Pool("pool1", offset="(0,0,0)", to="(conv1-east)"),
    to_Conv("conv2", 128, 64, offset="(1,0,0)", to="(pool1-east)", height=32, depth=32, width=2 ),
    to_connection( "pool1", "conv2"),
    to_Pool("pool2", offset="(0,0,0)", to="(conv2-east)", height=28, depth=28, width=1),
    to_SoftMax("soft1", 10 ,"(3,0,0)", "(pool1-east)", caption="SOFT"  ),
    to_connection("pool2", "soft1"),
    to_end()
    ]

def main():
    namefile = str(sys.argv[0]).split('.')[0]
    to_generate(arch, namefile + '.tex' )

if __name__ == '__main__':
    main()

Now, run the program as follows:

bash ../tikzmake.sh my_arch

plotneuralnet's Issues

Edge Thickness

Is it possible to change the thickness of the edges along a shape (such as a Box)? If so, how?

Thanks!

Labeling Blocks will create a solid line

I tried to let the Box block's edges be invisible via fill=\fill, draw=\fill in Box.sty.

Actually, It only let edges AH, AD, DE, CD, EF and HE invisible.

Finally I found that \path (b) edge ["\bf\xlabel"', midway] (c); created a solid line in BC, so do AB and CF.

I have no idea to fix this and I don't know why the solid lines emerged though no pic action acted in above code.

Furthermore, I want customize the style of draw line, but nothing changes after I put thin into "box" style setting.

How to draw a grid feature map

Thank you for your work and how to draw a grid feature map?

xdg-open instead of hard coding evince

Hey there,
great work !

I got just a small suggestion, the script ../tikzmake.sh hardcodes the evince pdf reader. However, not everybody uses that one.
Could I just change that like to:

#!/bin/bash


python $1.py 
pdflatex $1.tex

rm *.aux *.log *.vscodeLog
rm *.tex

xdg-open $1.pdf

In order to be a bit more compatible, even though thats a small issue

Zlabel Position Error

I just run fc8 from the example dir. I don't know why the position of Z-label is wrong ...

Suggestion for generating latex codes

It is an awesome project.
It is recommended to generate latex codes with pylatex, that makes the codes readable and extensible. Of course, you have to spend some time on it.

Is that possible to gen the pic from net config or vice-versa?

[Q]: How to do connection west - east ?

Unfortunately I'm not sure about the syntax in python.
And i don't want to change it every time in Latex.
\draw [connection] (dense_1-west) -- node {\midarrow} (reshape_1-east);

Thank you,
Vaclav

Closed

Closed

ResNet50 Tex

Will it be possible to get .tex for ResNet50?

When I am trying to define the architecture in python example: arch.py and then use the command "./tikzmake.sh arch" it is not working and giving me the following error:

Traceback (most recent call last):
File "arch.py", line 4, in
from pycore.tikzeng import *
ModuleNotFoundError: No module named 'pycore.tikzeng'
This is pdfTeX, Version 3.14159265-2.6-1.40.19 (TeX Live 2018) (preloaded format=pdflatex)
restricted \write18 enabled.
entering extended mode
! I can't find file `arch.tex'.
<*> arch.tex

(Press Enter to retry, or Control-D to exit)
Please type another input file name:

BatchNorm

Nice project! A batch normalization element would be cool...

[Can you push FCN8 code]

Thank you for your publish code. Can you add code of FCN-8 or shorcut connection code?

【help】HED

Thank you very much for your great work.
I want to draw the network structure of HED, but it has not been successful for a long time.
Can anyone share the original code of drawing HED? The deadline is coming. T-T
I'll be grateful.

Font size in the graph

Does anyone know how to change the font size in the graph? I tried to simply add to the documentclass

\documentclass[border=15pt, multi, tikz]{standalone}

but it doesn't work and returns some errors with positioning.

Suggested change to_ConvConvRelu

Hi. I was working with your code (great, by the way!) and, while using "to_ConvConvRelu", I realized that it could be easily be extended to any number of Convs:

def to_ConvConvRelu( name, s_filer=256, n_filer=(64,64), offset="(0,0,0)", to="(0,0,0)", width=(2,2), height=40, depth=40, caption=" " ):
    return r"""
\pic[shift={ """+ offset +""" }] at """+ to +""" 
    {RightBandedBox={
        name="""+ name +""",
        caption="""+ caption +""",
        xlabel={{ """+ ''.join([str(n_filer[i]) + """, """ for i in range(len(n_filer)-1)])+ str(n_filer[-1]) +""" }},
        zlabel="""+ str(s_filer) +""",
        fill=\ConvColor,
        bandfill=\ConvReluColor,
        height="""+ str(height) +""",
        width={ """+ ''.join([str(width[i]) + """, """ for i in range(len(width)-1)])+ str(width[-1]) +""" },
        depth="""+ str(depth) +"""
        }
    };
"""

Sorry for the not really clean code; there's surely a way of coding it that is not as dirty but it does the trick.

Regards,
Guille

Anyone would describe the requirements or dependencies?

Thanks to devs for developing this useful library.
However, I'm a pure layman to latex. Is there any version check for latex or some required python libraries?

TY.

Any preferred way to cite this repo?

Thanks for the awesome work!

Error when running readme example: `I can't find file `tikzlibraryquotes.code.tex'`

python my_arch.py works fine, producing the output included below. However, when I run pdflatex , I get error:

! I can't find file `tikzlibraryquotes.code.tex'.
<argument> ...nput tikzlibrary\pgf@temp .code.tex 
                                                  \catcode `\@ =\csname tikz...
l.2 \usetikzlibrary{quotes,arrows.meta}

I presume this is a simple fix requiring an install. Please may you help me diagnose what I need installed on my system for this to run? I am able to run tikz fine and create examples e.g. this one.

LaTeX info

$ pdflatex --version
pdfTeX 3.1415926-2.5-1.40.14 (TeX Live 2013/Debian)
kpathsea version 6.1.1
Copyright 2013 Peter Breitenlohner (eTeX)/Han The Thanh (pdfTeX).
There is NO warranty.  Redistribution of this software is
covered by the terms of both the pdfTeX copyright and
the Lesser GNU General Public License.
For more information about these matters, see the file
named COPYING and the pdfTeX source.
Primary author of pdfTeX: Peter Breitenlohner (eTeX)/Han The Thanh (pdfTeX).
Compiled with libpng 1.2.50; using libpng 1.2.50
Compiled with zlib 1.2.8; using zlib 1.2.8
Compiled with poppler version 0.24.5

Python output:

\documentclass[border=8pt, multi, tikz]{standalone} 
\usepackage{import}
\subimport{../layers/}{init}
\usetikzlibrary{positioning}
\usetikzlibrary{3d} %for including external image 


\def\ConvColor{rgb:yellow,5;red,2.5;white,5}
\def\ConvReluColor{rgb:yellow,5;red,5;white,5}
\def\PoolColor{rgb:red,1;black,0.3}
\def\UnpoolColor{rgb:blue,2;green,1;black,0.3}
\def\FcColor{rgb:blue,5;red,2.5;white,5}
\def\FcReluColor{rgb:blue,5;red,5;white,4}
\def\SoftmaxColor{rgb:magenta,5;black,7}   


\newcommand{\copymidarrow}{\tikz \draw[-Stealth,line width=0.8mm,draw={rgb:blue,4;red,1;green,1;black,3}] (-0.3,0) -- ++(0.3,0);}

\begin{document}
\begin{tikzpicture}
\tikzstyle{connection}=[ultra thick,every node/.style={sloped,allow upside down},draw=\edgecolor,opacity=0.7]
\tikzstyle{copyconnection}=[ultra thick,every node/.style={sloped,allow upside down},draw={rgb:blue,4;red,1;green,1;black,3},opacity=0.7]


\pic[shift={(0,0,0)}] at (0,0,0) 
    {Box={
        name=conv1,
        caption= ,
        xlabel={{64, }},
        zlabel=512,
        fill=\ConvColor,
        height=64,
        width=2,
        depth=64
        }
    };


\pic[shift={ (0,0,0) }] at (conv1-east) 
    {Box={
        name=pool1,
        caption= ,
        fill=\PoolColor,
        opacity=0.5,
        height=32,
        width=1,
        depth=32
        }
    };


\pic[shift={(1,0,0)}] at (pool1-east) 
    {Box={
        name=conv2,
        caption= ,
        xlabel={{64, }},
        zlabel=128,
        fill=\ConvColor,
        height=32,
        width=2,
        depth=32
        }
    };


\draw [connection]  (pool1-east)    -- node {\midarrow} (conv2-west);


\pic[shift={ (0,0,0) }] at (conv2-east) 
    {Box={
        name=pool2,
        caption= ,
        fill=\PoolColor,
        opacity=0.5,
        height=28,
        width=1,
        depth=28
        }
    };


\pic[shift={(3,0,0)}] at (pool1-east) 
    {Box={
        name=soft1,
        caption=SOFT,
        xlabel={{" ","dummy"}},
        zlabel=10,
        fill=\SoftmaxColor,
        opacity=0.8,
        height=3,
        width=1.5,
        depth=25
        }
    };


\draw [connection]  (pool2-east)    -- node {\midarrow} (soft1-west);


\end{tikzpicture}
\end{document}

what does height ,width and depth mean , in .tex file? how to change it?

In .tex file \pic[shift={(0,0,0)}] at (0,0,0) {RightBandedBox={name=cr1,caption=conv1,% xlabel={{"64","64"}},zlabel=I,fill=\ConvColor,bandfill=\ConvReluColor,% height=40,width={2,2},depth=40}}; but height ,width and depth ,how to initialize and change ?

Can you show us the python VGG16 code? Want to improve it for VGG19

Hello!

Thank you for your project, I love it!

I'm working on a VGG19 network and I want to update your code of VGG16 but I can't find the python code. Can you upload it please?

Kind regards

unet error

pyexamples|master⚡ ⇒ bash ../tikzmake.sh unet

File "unet.py", line 19
block_2ConvPool( name='b2', botton='pool_b1', top='pool_b2', s_filer=256, n_filer=128, offset="(1,0,0)", size=(32,32,3.5), opacity=0.5 ),
^
SyntaxError: invalid syntax
This is pdfTeX, Version 3.14159265-2.6-1.40.19 (TeX Live 2018) (preloaded format=pdflatex)
restricted \write18 enabled.
entering extended mode
! I can't find file `unet.tex'.
<> unet.tex

(Press Enter to retry, or Control-D to exit)
Please type another input file name: ^D
! Emergency stop.
<*> unet.tex

! ==> Fatal error occurred, no output PDF file produced!
Transcript written on texput.log.
rm: *.aux: No such file or directory
rm: *.vscodeLog: No such file or directory
rm: *.tex: No such file or directory
../tikzmake.sh: line 10: xdg-open: command not found

pdflatex + evince error

Hi awesome team,
I was trying to run example in the readme, was able to get the latex printed out in terminal but...

../tikzmake.sh: line 5: pdflatex: command not found
rm: *.aux: No such file or directory
rm: *.log: No such file or directory
rm: *.vscodeLog: No such file or directory
../tikzmake.sh: line 10: evince: command not found
line 5: pdflatex: command not found

Is there a way to not use these two?

How to draw a multiple inputs model?

I want to draw a model with 3 convolutional layer inputs and Integrated into the fully connected layer.How can I do this?

Unet to DenseNets

I am trying to modify Unet tex code to create DenseNets block.

I am wondering how to modify the connection lines that then look like densenets style?
Thank you for your time!

Is there a way to add legend?

Hi, I wonder if there is a way of adding legend to the figure (identifying which color corresponds to which type of layer).

A example would like the original U-Net figure:

Question about draw line ?

How can I darw the line like between upper point of a plane, not the center point between two plane?

Coordinate \name-top not defined in Box.sty

Apparently the Box.sty file is missing the \coordinate (\name-top) definition.

I want to know how to add legend

Hi, I wonder if there is a way of adding legend to the figure (identifying which color corresponds to which type of layer).

A example

And how to eliminate cube number?

Add a License

Hi,
first of all let me say the visualization you have created here is looking excellent! I was thinking about using it, but I have noticed that there is no license included in the project. Could you please add one? You could check out choosealicense.com if you are unsure which license fits your needs.
Thank you very much.

Any way to provide output image like using to_ouput()(like to_input())?

ModuleNotFoundError: No module named 'pycore.tikzeng'

After running
pip install pycore
when I try to run in python:
from pycore.tikzeng import *
I got:
ModuleNotFoundError: No module named 'pycore.tikzeng'

Version: pycore==17.4.17

Add a quick requirement for recent version of `pdflatex` at the top of your `README.md`

Referencing #26, worth guarding against silly issues like mine by simply insisting that the user has a recent version of pdflatex.

In my issue I show that 2013 pdflatex does not work, and 2018 does.

Error in readme example

In the examples section of the readme, the figure displayed is not VGG16, it is a FCN without the skip connections

subimport path error on windows

On windows, the path is connected by \ rather than /, which may cause some latex environment not recognize the subimport path correctly, e.g. on my machine (python 3.7 win10 x64), it produce

\subimport{..\layers/}{init}

but latexpdf complains about not find the path, it should be

\subimport{../layers/}{init}

if this can support model meta info ,ex caffe,tensorflow meta,very cool

What does the single quotation mark which follows double quotation marks mean?

Thanks for your amazing work!
I am confused with the quotation mark like "\cubezlabel"'. What is the difference between "\cubezlabel".

Looking forwarding your answer:)

May you add an example for U-Net?

May you add an example for U-Net architecture?
Thanks!

Transfer learning

Hi, is there a way to visualize transfer learning? means two parallel network and connections between them? or some hack in mind?

Thanks!

Cant get to_connection to work properly

arch = [
    to_head( '..' ),
    to_cor(),
    to_begin(),
    to_Conv("conv1", 320, 8, offset="(0,0,0)", to="(0,0,0)", height=64, depth=64, width=8),
    to_Pool("pool1", offset="(0,0,0)", to="(conv1-east)", height=60, depth=60, width=1),

    to_Conv("conv2", 160, 16, offset="(1,0,0)", to="(pool1-east)", height=32, depth=32, width=16),
    to_Pool("pool2", offset="(0,0,0)", to="(conv2-east)", height=30, depth=30, width=1),

    to_Conv("conv3", 80, 32, offset="(0,0,0)", to="(pool2-east)", height=16, depth=16, width=32),
    to_Pool("pool3", offset="(0,0,0)", to="(conv3-east)", height=1, depth=1, width=32, caption='GlobalAveragePooling'),

    to_SoftMax("soft1", 8 ,"(0,0,0)", "(pool3-east)", caption="fc1"),
    to_connection( "soft2", "soft1"),
    to_SoftMax("soft2", 1 ,"(1,0,0)", "(soft1-east)", caption="fc2", height=1, width=1, depth=1),
    to_end()
    ]

The above makes a green arrow through the whole network instead of the last 2 layers. Why?

citation

Hello
How do you want us to cite your work?

box edge colors

how to change the color of layer box from black to white or any other color. thanks.

Captioning the box will create a black dot only shown on Windows

Hi, @HarisIqbal88

I am back again 😸 ! I found that labeling a caption will make a black dot by the following code. Though it's invisible on my Mac OS and Ubuntu, I found it when I open the pdf on Windows, using Adobe reader exactly.

\path (\LastEastx/2,-\y/2,+\z/2) + (0,-25pt) coordinate (cap) 
        edge ["\textcolor{black}{ \bf \caption}"',captionlabel](cap) ; %Block caption/pic object label

I am wondering that the way you make labels by adding a edge, but the tutorials in TiKz always use node, which seems malfunctioned in your repo. (Maybe I use the node syntax wrongly)

Can you add some typical RNN or Transformer structures on ./example

Hi~ Thanks for your awesome contribution, it looks so cool for CNN related models!!!
but, can you give some examples like RNN/Stacked RNN/Transformer etc. really appreciate it

to_input(...) doesn't have east, west... properties

I would like to connect the input image to conv blocks and other blocks, but it is not possible since to_input(...) doesn't have east, west, northeast, etc... properties. How to solve this? Maybe wrap to_input(...) in a box or something?

Drawing only one conv layer

Hello,
thank you for this awesome project. If I draw only one conv layer like this:

\pic[shift={(0,0,0)}] at (0,0,0) {RightBandedBox={name=cr1,caption=conv1,%
        xlabel={{"96", ""}},ylabel=224,zlabel=224,fill=\ConvColor,bandfill=\ConvReluColor,%
        height=40,width={2},depth=40}};

then I have to keep xlabel as xlabel={{"96", ""}} with an empty qoutation marks. Is there a way to eliminate this need?

Problem when test example

Hi,everyone,when i follow the step in readme, I met a lot of problem.
I enter the name but it print the same thing.
It seems that I lack something. But I don't what is it. T^T

Way to change the fontsize

Hi, thank you for this awesome project! I am new to LaTex and just wondering is there any way for me to define or change the fontsize for the caption or the depthlabel? Thanks you!

Drawing the kernel dimensions in a conv layer

Hi,

Great library, very useful. I am writing a book for Cambridge University Press and would like to use it to explain CNNs. One feature that's missing is drawing the kernel dimensions (e.g. 3x3 or 5x5 or 11x11). Do you have any suggestions how to do that?

Michel

Importing layers in a latex project will affect captions on all figure boxes

Importing the 'layers' folder in a latex project will affect captions in every figure and table boxes. Basically, every caption will not have anymore the text "Figure X"/"Table X" where X is a index as soon as you add the lines \subimport{layers/}{init} .

Minimum example (correct):

\documentclass{article}

\usepackage{graphicx}
% \usepackage{tikz}
% \usepackage{import}
% \subimport{layers/}{init}
% \usetikzlibrary{positioning}

\begin{document}

\begin{figure}
    \centering
    \includegraphics[height=5cm]{cats.jpg}
    \caption{This is a caption.}
    \label{test2}
\end{figure}

\end{document}

This produces:

Importing layers:

\documentclass{article}

\usepackage{graphicx}
\usepackage{tikz}
\usepackage{import}
\subimport{layers/}{init}
\usetikzlibrary{positioning}

\begin{document}

\begin{figure}
    \centering
    \includegraphics[height=5cm]{cats.jpg}
    \caption{This is a caption.}
    \label{test2}
\end{figure}

\end{document}

Produces:

Issues so are:
not displaying anymore the count of captions and not keeping the correct alignment.

Python code for examples

Congratulations on a wonderful project.

Would you be able to provide the Python code for Unet_Ushape and the other examples for which it is not currently provided?

Thanks much -- Ian