sketch-nn 0.1.7

A tool to generate PyTorch neural network code from flowchart images

Sketch NN

Transform hand-drawn neural network sketches into functional PyTorch code

Sketch NN is an innovative Python library that transforms hand-drawn neural network sketches into functional PyTorch code. Design your neural architectures on paper, capture them with our tool, and watch as Sketch NN brings your ideas to life!

🌟 Features

• 📸 Process hand-drawn or digital sketches of neural network architectures
• 🧠 Support for a wide range of neural network layers
• 🔧 Generate ready-to-use PyTorch code
• 🖥️ User-friendly Gradio web interface for quick prototyping
• 🚀 FastAPI backend for scalable deployment

🛠️ Supported Layers

• Convolutional (Conv2D)
• Pooling (MaxPool2D, AvgPool2D)
• Fully Connected (Linear)
• Batch Normalization
• Dropout
• Activation Functions (ReLU, LeakyReLU, Sigmoid, Tanh)
• Recurrent (LSTM, GRU)
• Transformer
• Multi-head Attention

🚀 Installation

Install Sketch NN using pip:

pip install sketch_nn

🚀 Usage

from sketch_nn import NeuralNetworkDesigner

designer = NeuralNetworkDesigner()
designer.process_image('path_to_your_sketch.png')
pytorch_code = designer.generate_pytorch_code()
designer.write_to_file(pytorch_code, 'custom_nn.py')

Gradio Web Interface Demo

from sketch_nn.demo import run_gradio_demo

run_gradio_demo()