PlotNeuralNet is a Python package that provides tools to generate high-quality neural network architecture diagrams for research papers, presentations, and reports. It leverages LaTeX and Python for seamless integration into scientific workflows.

This package is based on the original PlotNeuralNet by HarisIqbal88, with improvements for better usability and a modular package structure.

• Programmatically generate neural network diagrams using Python.
• Predefined layer types (e.g., Conv, Pool, SoftMax).
• Easily extendable for custom layer shapes.
• Pre-built LaTeX templates for popular architectures like AlexNet, FCN, and HED.
• Fully structured as a Python package for streamlined integration into projects.

1. Clone the repository:

   git clone https://github.com/<your-username>/PlotNeuralNet.git
   cd PlotNeuralNet

2. Install the package:

   pip install .

3. Verify the installation:

   import PlotNeuralNet
   print("PlotNeuralNet installed successfully!")

The package is organized to simplify the creation of diagrams. It includes Python modules (pycore and pyexamples) and LaTeX resources (layers).

You can use the Python API to define your architecture programmatically. For example:

from PlotNeuralNet.pycore import tikzeng
from PlotNeuralNet.pycore.blocks import block_2ConvPool, block_Unconv

# Define architecture
arch = [
    tikzeng.to_head('..'),
    tikzeng.to_cor(),
    tikzeng.to_begin(),

    # Input image
    tikzeng.to_input('../examples/fcn8s/cats.jpg'),

    # Encoder
    *block_2ConvPool(name='b1', botton='input', top='b2', s_filer=256, n_filer=64),
    *block_2ConvPool(name='b2', botton='b2', top='b3', s_filer=128, n_filer=128),

    # Decoder
    *block_Unconv(name='b4', botton='b3', top='output', s_filer=64, n_filer=32),

    # Output layer
    tikzeng.to_ConvSoftMax(name='softmax', offset="(1,0,0)", to="(output-east)", width=1, height=30, depth=30),
    tikzeng.to_end(),
]

# Generate the architecture diagram
def main():
    tikzeng.to_generate(arch, "my_architecture.tex")

if __name__ == "__main__":
    main()

Run the Python script:

python my_architecture.py

Compile the .tex file with:

bash ../tikzmake.sh my_architecture

You can directly modify .tex files in the examples directory, such as examples/FCN-8 or examples/HED. Each .tex file demonstrates how to use LaTeX for defining architectures.

To compile a .tex file, use:

pdflatex <file>.tex

The package structure includes predefined resources for easy reuse:

• Available in the PlotNeuralNet/layers/ directory.
• Example LaTeX layer definitions:

  \input{layers/Box.sty}

• Predefined architectures like FCN, HED, AlexNet are in PlotNeuralNet/examples/.
• Modify these examples to fit your use case.

• Python examples for generating diagrams programmatically are in PlotNeuralNet/pyexamples/.
• Example usage:

  python PlotNeuralNet/pyexamples/unet.py

The package is organized as follows:

PlotNeuralNet/
├── LICENSE          # License file
├── MANIFEST.in      # File inclusion rules
├── README.md        # Documentation
├── setup.py         # Installation script
├── PlotNeuralNet/   # Main package directory
│   ├── __init__.py  # Package initializer
│   ├── pycore/      # Core Python modules
│   ├── layers/      # LaTeX resources
│   ├── examples/    # Predefined architectures in LaTeX
│   ├── pyexamples/  # Python examples
├── dist/            # Build artifacts (after running setup.py)
├── build/           # Temporary build files

1. Custom Layers

   • Extend pycore.blocks or create your own block definitions to support custom layers.

2. Batch Processing

   • Use Python scripts to generate multiple architectures programmatically.

3. Predefined Functions

   • block_2ConvPool and block_Unconv simplify common layer patterns.

This package is based on the original PlotNeuralNet by HarisIqbal88 and licensed under the MIT License.

This project is licensed under the MIT License. See the LICENSE file for more details.