[Tex/LaTex] Drawing a 1D Convolutional neural network

tikz-pgf

I am trying to show layers of a convolutional neural network like this picture:

I manage to draw the basic rectangle but how can I add details like this picture?

\documentclass[border=5pt, multi, tikz]{standalone}
\usetikzlibrary{quotes,arrows.meta}
\begin{document}
\begin{tikzpicture}[every edge quotes/.append style={auto, text=blue}]
  \pgfmathsetmacro{\cubex}{10}
  \pgfmathsetmacro{\cubey}{1}
  \pgfmathsetmacro{\cubez}{1}
  \draw [draw=blue, every edge/.append style={draw=blue, densely dashed, opacity=.5}, fill=none]
    (0,0,0) coordinate (o) -- ++(-\cubex,0,0) coordinate (a) -- ++(0,-\cubey,0) coordinate (b) edge coordinate [pos=1] (g) ++(0,0,-\cubez)  -- ++(\cubex,0,0) coordinate (c) -- cycle
    (o) -- ++(0,0,-\cubez) coordinate (d) -- ++(0,-\cubey,0) coordinate (e) edge (g) -- (c) -- cycle
    (o) -- (a) -- ++(0,0,-\cubez) coordinate (f) edge (g) -- (d) -- cycle;
  \path [every edge/.append style={draw=blue, |-|}]
    (b) +(0,-5pt) coordinate (b1) edge ["500"'] (b1 -| c)
    (b) +(-5pt,0) coordinate (b2) (b2 |- a)
    (c) +(3.5pt,-3.5pt) coordinate (c2) edge ["1"'] ([xshift=3.5pt,yshift=-3.5pt]e)
    ;
\end{tikzpicture}
\end{document}

Best Answer

This is to give you a start. This answer shows how to stack the boxes in 3D, and to draw connections between corners and so on. It does, however, not attempt to reproduce all details of your screen shot. However, if you have conceptual problems, i.e. ask for features that go beyond repetition, I will be happy to update my answer.

\documentclass[tikz,border=3.14mm]{standalone}
\usepackage{tikz-3dplot}
\usetikzlibrary{calc}
\tikzset{pics/fake box/.style args={% #1=color, #2=x dimension, #3=y dimension, #4=z dimension
#1 with dimensions #2 and #3 and #4}{
code={
\draw[fill=#1!50]  (0,#3,0) coordinate(-front-bottom-right) to
 (0,#3,#4) coordinate(-front-top-right) --
(-#2,#3,#4) coordinate(-front-top-left) -- (-#2,#3,0) 
coordinate(-front-bottom-left) -- cycle;
\draw[fill=#1] (0,0,#4)  -- (-#2,0,#4) 
-- (-#2,#3,#4) coordinate(-back-top-left) -- (0,#3,#4) 
 coordinate(-back-top-right)  -- cycle;
\draw[fill=#1] (-#2,0,0) coordinate(-back-bottom-left) --
(-#2,0,#4) -- (-#2,#3,#4) -- (-#2,#3,0)
-- cycle;
}},
pics/cake box/.style args={% #1=color, #2=x dimension, #3=y dimension, #4=z dimension
#1 with dimensions #2 and #3 and #4 and label #5}{
code={
\draw[#1]  (0,#3,0) coordinate(-front-bottom-right) to
 (0,#3,#4) coordinate(-front-top-right) -- 
(-#2,#3,#4) coordinate(-front-top-left) -- (-#2,#3,0) 
coordinate(-front-bottom-left) -- cycle;
\draw[#1] (0,0,#4)  -- (-#2,0,#4) 
-- (-#2,#3,#4) coordinate(-back-top-left) -- 
node[midway,above=3pt,sloped,text=black]{#5} (0,#3,#4) 
 coordinate(-back-top-right)  -- cycle;
\draw[#1] (-#2,0,0) coordinate(-back-bottom-left) --
(-#2,0,#4) -- (-#2,#3,#4) -- (-#2,#3,0)
-- cycle;
}},
}
% from https://tex.stackexchange.com/a/52856/121799
\tikzset{circle dotted/.style={dash pattern=on .05mm off 2mm,
                                         line cap=round}}
\begin{document}
\tdplotsetmaincoords{110}{20}
\begin{tikzpicture}[tdplot_main_coords,transform shape,
font=\sffamily\small,line join=bevel]
%
% comment these out if you want to see where the axes point to
% \draw[-latex] (0,0,0) -- (3,0,0) node[below]{$x$};
% \draw[-latex] (0,0,0) -- (0,3,0) node[left]{$y$};
% \draw[-latex] (0,0,0) -- (0,0,3) node[below]{$z$};
% a plane
\foreach \X/\Y [count=\Z starting from 0] in {10/1,9/3,7/3,6/5}
{
\draw pic (box1-\Z) at (0.3*\Z,0,-2*\Z) 
{cake box={red,thin} with dimensions {\X*0.4} and {\Y*0.1} and {0.2} and label
{Layer \Z}};
}
% add boxes left of lowest main box (Layer 3)
\path pic (box2-3) at (box1-3-back-bottom-left) 
{fake box={blue} with dimensions {0.3} and {5*0.1} and 0.2};
\path pic (box3-3) at (box2-3-back-bottom-left) 
{cake box={red} with dimensions {0.5} and {5*0.1} and 0.2 and label {}};
\path pic (box4-3) at (box3-3-back-bottom-left) 
{fake box={green} with dimensions {0.1} and {5*0.1} and 0.2};
\path pic (box5-3) at (box4-3-back-bottom-left) 
{cake box={red} with dimensions {0.3} and {5*0.1} and 0.2 and label {}};
% add boxes left of next-to-lowest main box (Layer 2)
\path pic (box2-2) at (box1-2-back-bottom-left) 
{fake box={green} with dimensions {0.3} and {3*0.1} and 0.2};
\path pic (box3-2) at (box2-2-back-bottom-left) 
{cake box={red} with dimensions {0.5} and {3*0.1} and 0.2 and label {}};
\path pic (box4-2) at (box3-2-back-bottom-left) 
{fake box={blue} with dimensions {0.1} and {3*0.1} and 0.2};
\path pic (box5-2) at (box4-2-back-bottom-left) 
{cake box={red} with dimensions {0.3} and {3*0.1} and 0.2 and label {}};
%
% and so on 
%
% connections
\draw[gray] (box2-2-front-bottom-right) -- (box4-3-front-top-right);
\draw[gray] (box2-2-front-bottom-left) -- (box4-3-back-top-left);
% bars
\tikzset{indicate dimension/.style args={from #1 to #2 with #3}{
insert path={($#1!1mm!-90:#2$) -- ($#1!3mm!-90:#2$)
 ($#2!1mm!90:#1$) -- ($#2!3mm!90:#1$)
 ($#1!2mm!-90:#2$) -- ($#2!2mm!90:#1$)
 node[midway,below,sloped]{#3}}}}
\draw[very thin,gray,indicate dimension={from (box5-3-front-bottom-left) to (box1-3-front-bottom-right) with 244}];
\draw[very thin,gray,indicate dimension={from (box5-2-front-bottom-left) to (box1-2-front-bottom-right) with 248}];
\end{tikzpicture}
\end{document}

Related Solutions

[Tex/LaTex] Rotate a node but not its content: the case of the ellipse decoration

The source of the difficulty is that ellipses are constructed in a particular way in TikZ. They are paths that start from the x-axis and proceed counter-clockwise around their centre. The vast majority of the time, the exact parametrisation doesn't matter. You appear to have found the one situation where it does!

In the actual question, you only want to be able to mirror the ellipse, and so draw it starting from the negative x-axis (the title of the question suggests a more flexible approach). That's actually not too hard since we can exploit the symmetry of the ellipse. The key is to provide it with a negative x-radius, since then it will start from the negative x-axis (and proceed clockwise, but we could correct for that by negating the y-radius as well). To do this, we interrupt the call from the node shape to the drawing command and flip the sign of the x-radius. The simplest way to do this is to redefine the \pgfpathellipse macro to do the negation and then call the original macro. The following code does this.

\documentclass{article}
\usepackage{tikz}
\usetikzlibrary{decorations,shapes,decorations.markings}
\makeatletter
\let\origpgfpathellipse=\pgfpathellipse
\def\revpgfpathellipse#1#2#3{%
  #2%
  \pgf@xa=-\pgf@x
  \origpgfpathellipse{#1}{\pgfqpoint{\pgf@xa}{0pt}}{#3}}
\makeatother

\tikzset{
  reversed ellipse/.style={
    ellipse,
    reverse the ellipse%
  },
  reverse the ellipse/.code={
    \let\pgfpathellipse=\revpgfpathellipse
  }
}
\begin{document}
\begin{tikzpicture}
\node[ellipse,
  draw,
  postaction={
    decorate,
    decoration={
      markings,
      mark=at position 1 with {
        \arrow[line width=5pt,blue]{>}
      }
    }
  }
] at (0,0) {hello world};

\node[reversed ellipse,
  draw,
  postaction={
    decorate,
    decoration={
      markings,
      mark=at position 1 with {
        \arrow[line width=5pt,blue]{>}
      }
    }
  }
] at (0,-2) {hello world};
\end{tikzpicture}

\end{document}

Here's the result:

rotated ellipse

(the arrow got clipped, but you can see where it lies)

[Tex/LaTex] Drawing neural network with tikz

Here, we a have a festival of \foreach:

\documentclass[border=0.125cm]{standalone}
\usepackage{tikz}
\usetikzlibrary{positioning}
\begin{document}

\tikzset{%
  every neuron/.style={
    circle,
    draw,
    minimum size=1cm
  },
  neuron missing/.style={
    draw=none, 
    scale=4,
    text height=0.333cm,
    execute at begin node=\color{black}$\vdots$
  },
}

\begin{tikzpicture}[x=1.5cm, y=1.5cm, >=stealth]

\foreach \m/\l [count=\y] in {1,2,3,missing,4}
  \node [every neuron/.try, neuron \m/.try] (input-\m) at (0,2.5-\y) {};

\foreach \m [count=\y] in {1,missing,2}
  \node [every neuron/.try, neuron \m/.try ] (hidden-\m) at (2,2-\y*1.25) {};

\foreach \m [count=\y] in {1,missing,2}
  \node [every neuron/.try, neuron \m/.try ] (output-\m) at (4,1.5-\y) {};

\foreach \l [count=\i] in {1,2,3,n}
  \draw [<-] (input-\i) -- ++(-1,0)
    node [above, midway] {$I_\l$};

\foreach \l [count=\i] in {1,n}
  \node [above] at (hidden-\i.north) {$H_\l$};

\foreach \l [count=\i] in {1,n}
  \draw [->] (output-\i) -- ++(1,0)
    node [above, midway] {$O_\l$};

\foreach \i in {1,...,4}
  \foreach \j in {1,...,2}
    \draw [->] (input-\i) -- (hidden-\j);

\foreach \i in {1,...,2}
  \foreach \j in {1,...,2}
    \draw [->] (hidden-\i) -- (output-\j);

\foreach \l [count=\x from 0] in {Input, Hidden, Ouput}
  \node [align=center, above] at (\x*2,2) {\l \\ layer};

\end{tikzpicture}

\end{document}

Although it seems unwise to have n denote the number of nodes in each layer when they could be different and the arrangement of the diagram suggests they are not.

enter image description here

Best Answer

Related Solutions

[Tex/LaTex] Rotate a node but not its content: the case of the ellipse decoration

[Tex/LaTex] Drawing neural network with tikz

Related Question