Skip to content

Chapter 35: TensorFlow Lite & Mobile Deployment

“A model isn’t truly intelligent until it fits in your pocket.”

📱 Introduction: Why Deploy on Mobile?

TensorFlow Lite (TFLite) is TensorFlow’s lightweight version tailored for mobile and edge devices like Android, iOS, Raspberry Pi, and microcontrollers. With the rise of real-time AI on phones (face unlock, voice assistants, image classifiers), deploying efficient models to run locally is now a core part of ML engineering.

In this chapter, you’ll learn how to:

  • Convert TensorFlow models to .tflite
  • Optimize for performance with quantization
  • Deploy and test on Android/iOS
  • Run inference in Python or embedded systems

Step 1: Convert a TensorFlow Model to TFLite

We’ll begin with a simple image classifier (e.g., trained on MNIST or MobileNet). 

import tensorflow as tf

# Load or train a model (example: MobileNetV2 pretrained)
model = tf.keras.applications.MobileNetV2(weights="imagenet", input_shape=(224, 224, 3))

# Convert to TFLite
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()

# Save the model
with open("mobilenetv2.tflite", "wb") as f:
    f.write(tflite_model)

Step 2: Optimize the Model with Quantization

Reducing model size and improving speed, especially for devices with limited resources. 

# Dynamic Range Quantization (fastest + simplest)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
quantized_model = converter.convert()

with open("mobilenetv2_quant.tflite", "wb") as f:
    f.write(quantized_model)
Other options:

  • Post-training integer quantization

  • Full integer quantization (with representative dataset)

  • Float16 quantization for GPUs

Step 3: Run Inference with TFLite Interpreter (Python)

import numpy as np
from PIL import Image

# Load TFLite model
interpreter = tf.lite.Interpreter(model_path="mobilenetv2_quant.tflite")
interpreter.allocate_tensors()

# Get input/output details
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()

# Prepare image input
img = Image.open("cat.jpg").resize((224, 224))
input_data = np.expand_dims(np.array(img, dtype=np.float32) / 255.0, axis=0)

# Run inference
interpreter.set_tensor(input_details[0]['index'], input_data)
interpreter.invoke()
output_data = interpreter.get_tensor(output_details[0]['index'])

print("Prediction:", np.argmax(output_data))

📲 Step 4: Deploy to Android/iOS

You can now take your .tflite model and integrate it into:

  • Android Studio using TensorFlow Lite Task Library or Interpreter API

  • iOS apps using Swift or Objective-C

  • Flutter apps using the tflite_flutter plugin

💡 Bonus: Use ML Model Binding in Android Studio for no-code integration of .tflite models with image or text inputs!

Other Deployment Targets

TensorFlow Lite isn’t just for phones:

Platform Use Case
Raspberry Pi Smart cameras, IoT vision
Coral Edge TPU Ultra-fast inference with acceleration
Microcontrollers TinyML with TensorFlow Lite for Microcontrollers

Summary

In this chapter, you learned:

  • How to convert Keras models into efficient .tflite files

  • How to optimize models with quantization

  • How to run inference in Python, Android, or iOS

  • That TFLite opens the door to AI at the edge — fast, offline, and private

TensorFlow Lite brings ML to environments where connectivity and power are limited, but the need for real-time intelligence is high — from smartwatches and cameras to drones and industrial machines.