TensorFlow.js conversion

Introduction

TensorFlow.js conversion lets you use machine learning models directly in a web browser. This means you can run AI without needing a server.

You want to run a trained TensorFlow model inside a website.

You want users to get instant AI results without internet delays.

You want to share your AI model easily without complex setup.

You want to build interactive AI demos that run on any device.

You want to protect user data by running AI locally in the browser.

Syntax

TensorFlow

tensorflowjs_converter \
  --input_format=tf_saved_model \
  /path/to/saved_model \
  /path/to/tfjs_target_dir

The command converts a TensorFlow SavedModel to TensorFlow.js format.

You can also convert Keras models or frozen graphs by changing --input_format.

Examples

Convert a TensorFlow SavedModel folder my_model to TensorFlow.js format in tfjs_model folder.

TensorFlow

tensorflowjs_converter --input_format=tf_saved_model ./my_model ./tfjs_model

Convert a Keras H5 model file model.h5 to TensorFlow.js format.

TensorFlow

tensorflowjs_converter --input_format=keras ./model.h5 ./tfjs_keras_model

Convert a frozen TensorFlow graph to TensorFlow.js format, specifying output nodes.

TensorFlow

tensorflowjs_converter --input_format=tf_frozen_model --output_node_names='output_node' ./frozen_model.pb ./tfjs_frozen_model

Sample Model

This code creates a simple neural network, saves it, and converts it to TensorFlow.js format using the command line tool inside Python.

TensorFlow

import tensorflow as tf
import tensorflowjs as tfjs

# Create a simple Keras model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, activation='relu', input_shape=(5,)),
    tf.keras.layers.Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# Save the model in TensorFlow SavedModel format
model.save('saved_model')

# Convert the saved model to TensorFlow.js format
import subprocess
subprocess.run([
    'tensorflowjs_converter',
    '--input_format=tf_saved_model',
    'saved_model',
    'tfjs_model'
])

print('Model converted to TensorFlow.js format in folder tfjs_model')

Important Notes

Make sure you have tensorflowjs installed: pip install tensorflowjs.

The conversion creates files that can be loaded in JavaScript with tf.loadLayersModel().

Check the TensorFlow.js documentation for advanced options like quantization.

Summary

TensorFlow.js conversion lets you run TensorFlow models in web browsers.

You convert models using the tensorflowjs_converter tool.

Converted models can be loaded easily in JavaScript for fast, local AI.