visualkeras for Keras / TensorFlow

Visualkeras is a Python package for visualizing Keras and TensorFlow model architectures. It supports several rendering styles, such as classic layered CNN diagrams, node-based visualizations, and LeNet-style visualizations. It is very easy to get started with visualkeras (see Quickstart), but also highly customizable for advanced users. For help in citing this project, refer here.

Installation

Install the latest published release:

pip install visualkeras

Install the latest master branch (potentially unstable):

pip install git+https://github.com/paulgavrikov/visualkeras

Quick Start

import visualkeras

model = ...

visualkeras.layered_view(model).show()
visualkeras.layered_view(model, to_file="model.png")

The recommended high-level API is show(...), which selects a renderer by mode:

import visualkeras
from tensorflow.keras import layers
from visualkeras.options import FunctionalOptions

img = visualkeras.show(
    model,
    mode="functional",
    options=FunctionalOptions(
        collapse_enabled=True,
        collapse_rules=[
            {"kind": "layer", "selector": layers.Dense, "repeat_count": 4},
            {
                "kind": "block",
                "selector": [layers.Dense, layers.Dropout],
                "repeat_count": 2,
                "annotation_position": "below",
            },
        ],
    ),
)

show(...) supports these modes:

• layered
• graph
• functional
• lenet

Renderers

Examples

We provide basic examples here. Explamples with various options and customizations are covered in the documentation: .

Layered view

import tensorflow as tf
from tensorflow import keras
import visualkeras

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"),
])

visualkeras.layered_view(model).show()

Graph view

import tensorflow as tf
from tensorflow import keras
import visualkeras

model = Sequential([
    Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    MaxPooling2D((2, 2)),
    Dense(10, activation='softmax')
])

visualkeras.graph_view(model)

Functional view

import tensorflow as tf
from tensorflow import keras
import visualkeras

inputs = keras.Input(shape=(16,))
x = keras.layers.Dense(32, activation='relu')(inputs)
x = keras.layers.Dense(32, activation='relu')(x)
outputs = keras.layers.Dense(10, activation='softmax')(x)

model = keras.Model(inputs=inputs, outputs=outputs)

visualkeras.functional_view(model)

LeNet view

import tensorflow as tf
from tensorflow import keras
import visualkeras

model = keras.Sequential([
    keras.layers.Input(shape=(28, 28, 1)),
    keras.layers.Conv2D(6, (5, 5), activation='tanh'),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(16, (5, 5), activation='tanh'),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Flatten(),
    keras.layers.Dense(120, activation='tanh'),
    keras.layers.Dense(84, activation='tanh'),
    keras.layers.Dense(10, activation='softmax')
])

visualkeras.lenet_view(model)

Documentation

Detailed documentation can be found in the documentation website: .

Particularly useful sections include:

• Quickstart:
• Tutorials:
• Examples:
• API reference:
• Architecture guide: ARCHITECTURE.md

Compatibility

Citation

If you find this project helpful for your research, please cite it:

@misc{Gavrikov2020VisualKeras,
  author = {Gavrikov, Paul and Patapati, Santosh},
  title = {visualkeras},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/paulgavrikov/visualkeras}},
}

Contributing

• Issues:
• Contributing guide: CONTRIBUTING.MD
• Architecture guide: ARCHITECTURE.md

Stay Updated

Sign up for the visualkeras mailing list here:

License

Visualkeras is licensed under the MIT License. See LICENSE.