Mobile deployment (TFLite, Core ML) in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - Mobile deployment (TFLite, Core ML)

Which metric matters for Mobile deployment (TFLite, Core ML) and WHY

When deploying computer vision models on mobile devices using TFLite or Core ML, the key metrics to focus on are model size, inference speed, and accuracy.

Model size matters because mobile devices have limited storage and memory. Smaller models load faster and use less space.

Inference speed is important because users expect quick responses. A slow model leads to poor user experience.

Accuracy remains critical to ensure the model makes correct predictions despite being smaller or faster.

Balancing these metrics ensures the model works well on mobile without draining battery or causing delays.

Confusion matrix example for a mobile image classifier

      Actual \ Predicted | Cat | Dog | Bird
      -------------------------------------
      Cat                | 45  | 3   | 2
      Dog                | 4   | 40  | 6
      Bird               | 1   | 5   | 44
    

This matrix shows how many images of each animal were correctly or incorrectly classified. It helps measure accuracy, precision, and recall for each class.

Precision vs Recall tradeoff in mobile deployment

Imagine a mobile app that detects if a photo contains a dog.

• High precision means when the app says "dog," it is usually right. This avoids false alarms but might miss some dogs.
• High recall means the app finds most dogs in photos but might sometimes say "dog" when there isn't one.

On mobile, you might prefer high precision to avoid annoying users with wrong alerts, even if some dogs are missed.

But if missing a dog is critical (like a safety app), you might choose high recall.

What good vs bad metrics look like for mobile deployment

• Good: Model size under 10MB, inference time under 100ms, accuracy above 85%
• Bad: Model size over 50MB (slow to load), inference time over 1 second (laggy), accuracy below 70% (many wrong predictions)

Good metrics mean the app feels fast, uses little space, and predicts well. Bad metrics cause slow, clunky apps with poor results.

Common pitfalls in mobile deployment metrics

• Ignoring latency: A model with high accuracy but slow inference is bad for mobile.
• Overfitting: A model that works great on test data but poorly on real mobile images.
• Data leakage: Training data too similar to test data inflates accuracy falsely.
• Ignoring battery use: Complex models drain battery quickly.
• Not testing on real devices: Metrics from desktop may not reflect mobile performance.

Self-check question

Your mobile model has 98% accuracy but takes 2 seconds to classify one image. Is it good for production? Why or why not?

Answer: No, it is not good. Even with high accuracy, 2 seconds per image is too slow for a smooth mobile experience. Users expect quick results, so inference speed must improve.

Key Result

For mobile deployment, balancing small model size, fast inference, and good accuracy is key to a good user experience.

Practice

1. What is the main purpose of using TFLite or Core ML in mobile deployment?

easy

A. To replace mobile operating systems with AI-powered ones

B. To run AI models directly on mobile devices for faster and offline use

C. To collect data from mobile devices for training

D. To train AI models on mobile devices

Solution

1. Step 1: Understand mobile deployment goals

   Mobile deployment aims to run AI models on phones to improve speed and allow offline use.

2. Step 2: Identify TFLite and Core ML roles

   TFLite and Core ML are formats to convert models for running directly on Android and Apple devices respectively.

3. Final Answer:

   To run AI models directly on mobile devices for faster and offline use -> Option B

4. Quick Check:

   Mobile AI models run locally = D [OK]

Hint: Mobile AI runs on device for speed and offline use [OK]

Common Mistakes:

• Thinking TFLite/Core ML train models on phones
• Confusing data collection with deployment
• Assuming they replace mobile OS

2. Which of the following is the correct command to convert a TensorFlow model to TFLite format in Python?

easy

A. tflite_model = tf.convert_to_tflite('model_dir')

B. tflite_model = tf.saved_model.convert_to_tflite('model_dir')

C. tflite_model = tf.lite.convert('model_dir')

D. tflite_model = tf.lite.TFLiteConverter.from_saved_model('model_dir').convert()

Solution

1. Step 1: Recall TensorFlow Lite conversion syntax

   The official way is using tf.lite.TFLiteConverter.from_saved_model() to load and convert.

2. Step 2: Check each option's correctness

   Only tflite_model = tf.lite.TFLiteConverter.from_saved_model('model_dir').convert() uses the correct method and chaining to convert the model.

3. Final Answer:

   tflite_model = tf.lite.TFLiteConverter.from_saved_model('model_dir').convert() -> Option D

4. Quick Check:

   Use tf.lite.TFLiteConverter.from_saved_model() = B [OK]

Hint: Use tf.lite.TFLiteConverter.from_saved_model() to convert [OK]

Common Mistakes:

• Using non-existent tf.convert_to_tflite function
• Calling convert() on wrong object
• Mixing saved_model and convert_to_tflite methods

3. Given the following Python code snippet, what will be the output type of tflite_model after conversion?

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

medium

A. A string path to the converted model file

B. A TensorFlow SavedModel object

C. A bytes object containing the TFLite model

D. A Python dictionary with model details

Solution

1. Step 1: Understand the convert() method output

   The convert() method returns a bytes object representing the TFLite flatbuffer model.

2. Step 2: Match output type to options

   Only A bytes object containing the TFLite model correctly states the output is a bytes object containing the TFLite model.

3. Final Answer:

   A bytes object containing the TFLite model -> Option C

4. Quick Check:

   convert() returns bytes = A [OK]

Hint: convert() returns bytes of TFLite model, not file path [OK]

Common Mistakes:

• Thinking convert() saves file automatically
• Expecting a model object instead of bytes
• Confusing output with string path

4. You tried to convert a Core ML model using the command coremltools.converters.convert('model.mlmodel') but got an error. What is the likely cause?

medium

A. The convert function requires a model object, not a file path string

B. The model file extension must be .tflite for Core ML conversion

C. Core ML models cannot be converted with coremltools

D. The convert function only works on TensorFlow models

Solution

1. Step 1: Understand coremltools convert function input

   The convert function expects a model object or supported format, not just a file path string.

2. Step 2: Identify the error cause

   Passing a string path directly causes an error because the function cannot load the model from string alone.

3. Final Answer:

   The convert function requires a model object, not a file path string -> Option A

4. Quick Check:

   convert() needs model object input = C [OK]

Hint: Pass model object, not file path string, to convert() [OK]

Common Mistakes:

• Confusing file extensions for Core ML
• Thinking coremltools can't convert Core ML models
• Assuming convert() only works on TensorFlow

5. You have a trained TensorFlow model and want to deploy it on both Android and iOS devices. Which sequence of steps correctly prepares the model for mobile deployment?

hard

A. Convert the TensorFlow model to TFLite format for Android, then convert the same TensorFlow model to Core ML format for iOS

B. Convert the TensorFlow model to Core ML format for Android, then convert to TFLite for iOS

C. Use the TensorFlow model directly on both Android and iOS without conversion

D. Convert the TensorFlow model to ONNX format, then use ONNX runtime on both Android and iOS

Solution

1. Step 1: Identify platform-specific model formats

   Android uses TFLite format, and iOS uses Core ML format for efficient mobile deployment.

2. Step 2: Convert TensorFlow model accordingly

   Convert the TensorFlow model separately to TFLite for Android and Core ML for iOS to ensure compatibility.

3. Final Answer:

   Convert the TensorFlow model to TFLite format for Android, then convert the same TensorFlow model to Core ML format for iOS -> Option A

4. Quick Check:

   Platform-specific formats: TFLite for Android, Core ML for iOS = A [OK]

Hint: Convert TensorFlow model separately for Android (TFLite) and iOS (Core ML) [OK]

Common Mistakes:

• Mixing Core ML format for Android devices
• Skipping conversion and using TensorFlow model directly
• Using ONNX runtime without proper support