Tutorial 1: Basic Visualization#

Estimated time: 15 minutes

In this tutorial, you’ll learn the basics of visualizing neural network architectures using visualkeras.

Overview#

By the end of this tutorial, you’ll be able to:

  • Understand the difference between layered view and graph view

  • Create your first visualization

  • Know when to use each visualization style

  • Customize basic appearance

Prerequisites#

  • visualkeras installed (see Installation)

  • Basic knowledge of Keras/TensorFlow models

  • Jupyter notebook or Python IDE

Visualization Modes#

visualkeras provides four visualization modes, all accessed through the unified show() API:

mode=’layered’ — Stacked layers

  • Shows layers as 3D blocks stacked vertically

  • Great for CNNs and image processing networks

  • Emphasizes tensor dimensions flowing through layers

  • Best for understanding layer-by-layer transformations

mode=’graph’ — Computational graph

  • Shows model as a connected graph

  • Works with any model type (Sequential, Functional, Subclassed)

  • Clear visualization of connections and branching

  • Best for complex models with multiple paths

mode=’functional’ — Graph-aware layering

  • Combines best of both: layered appearance with graph awareness

  • Shows connections while maintaining layer structure

  • Ideal for CNNs with skip connections or multi-branch architectures

  • Supports collapse rules to merge repeated layer sequences

mode=’lenet’ — Feature map stack

  • Classic “feature map stacks” visualization style

  • Shows channels as individual map visualizations

  • Perfect for publishing CNN architectures in papers

  • Ideal for detailed architectural diagrams

Creating Your First Visualization#

Let’s start with a simple CNN and see all four modes:

import tensorflow as tf
from tensorflow import keras
import visualkeras

# Define a simple CNN
model = keras.Sequential([
    keras.layers.Input(shape=(28, 28, 1)),
    keras.layers.Conv2D(32, (3, 3), activation='relu'),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(64, (3, 3), activation='relu'),
    keras.layers.Flatten(),
    keras.layers.Dense(10, activation='softmax')
])

# The unified show() API with mode parameter
# Try different modes to see what works best for your use case

# Layered view
image = visualkeras.show(model, mode='layered')
image.show()

The unified show() API makes it easy to try different visualization styles without changing function calls.

Comparing All Four Modes#

Here’s how to visualize your model in all four modes:

import visualkeras

# Try each visualization mode
layered = visualkeras.show(model, mode='layered')
graph = visualkeras.show(model, mode='graph')
functional = visualkeras.show(model, mode='functional')
lenet = visualkeras.show(model, mode='lenet')

Each mode reveals different aspects of your architecture. Experiment to find what works best for your presentation or publication.

When to Use Each Mode#

Use Layered when:

  • Visualizing CNNs or image processing networks

  • Want to emphasize layer stacking and tensor shapes

  • Have relatively simple sequential architectures

  • Presenting to people new to neural networks

Use Graph when:

  • Working with complex models with many branches

  • Have skip connections or parallel layers

  • Want the most compact representation

  • Visualizing Functional or Subclassed models

Use Functional when:

  • Have a CNN with non-sequential elements (skip connections)

  • Want layered appearance but with graph connections shown

  • Need something between pure layered and pure graph

  • Working with complex feature extraction pipelines

Use LeNet when:

  • Publishing architectures in academic papers

  • Want classic “feature maps on paper” style

  • Need detailed per-channel visualization

  • Designing presentation slides with detailed diagrams

Common Issues#

Blank visualization

Make sure your model has an Input layer:

# With Input layer (works)
model = keras.Sequential([
    keras.layers.Input(shape=(28, 28, 1)),
    keras.layers.Conv2D(32, (3, 3)),
])
Tensor shapes not showing

Build the model explicitly:

model.build((None, 28, 28, 1))
visualkeras.layered_view(model).show()

Next Steps#

Now you understand the basics! Move on to: