Mobile deployment (TFLite, Core ML) in Computer Vision

Introduction

Mobile deployment lets you run AI models directly on phones. This makes apps faster and works without internet.

You want a photo app that recognizes objects instantly on your phone.

You need a voice assistant that works offline on mobile devices.

You want to save data by processing AI tasks on the phone instead of the cloud.

You are building a fitness app that tracks movements using the phone camera.

You want to add AI features to an app without slowing it down.

Syntax

Computer Vision

For TensorFlow Lite (TFLite):

import tensorflow as tf

# Convert a TensorFlow model to TFLite format
converter = tf.lite.TFLiteConverter.from_saved_model('saved_model_dir')
tflite_model = converter.convert()

# Save the TFLite model
with open('model.tflite', 'wb') as f:
    f.write(tflite_model)

For Core ML (Apple devices):

import coremltools as ct

# Convert a TensorFlow or PyTorch model to Core ML
coreml_model = ct.convert(model)

# Save the Core ML model
coreml_model.save('model.mlmodel')

TFLite is for Android and many platforms; Core ML is for Apple devices.

Conversion changes the model to a smaller, faster format for mobile.

Examples

This converts a saved TensorFlow model to TFLite format and saves it.

Computer Vision

import tensorflow as tf

converter = tf.lite.TFLiteConverter.from_saved_model('my_model')
tflite_model = converter.convert()

with open('model.tflite', 'wb') as f:
    f.write(tflite_model)

This converts a TensorFlow model to Core ML format for iOS apps.

Computer Vision

import coremltools as ct

coreml_model = ct.convert(my_tf_model)
coreml_model.save('MyModel.mlmodel')

Sample Model

This code trains a small model, saves it, converts it to TFLite, and prints the TFLite model size.

Computer Vision

import tensorflow as tf
import numpy as np

# Create a simple TensorFlow model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(3, activation='relu', input_shape=(2,)),
    tf.keras.layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# Train on dummy data
x_train = np.array([[0,0],[0,1],[1,0],[1,1]], dtype=np.float32)
y_train = np.array([0,1,1,0], dtype=np.float32)
model.fit(x_train, y_train, epochs=5, verbose=1)

# Save the model
model.save('saved_model')

# Convert to TFLite
converter = tf.lite.TFLiteConverter.from_saved_model('saved_model')
tflite_model = converter.convert()

with open('model.tflite', 'wb') as f:
    f.write(tflite_model)

print('TFLite model size:', len(tflite_model), 'bytes')

Important Notes

Always test the converted model on the target device to check speed and accuracy.

Some model features may not convert perfectly; keep models simple for mobile.

Use quantization during conversion to make models smaller and faster.

Summary

Mobile deployment runs AI models directly on phones for speed and offline use.

TFLite is for Android and Core ML is for Apple devices.

Convert your trained model to these formats before adding to mobile apps.