Image Classification for Emotion Recognition

This project demonstrates how to build an image classification model for emotion recognition using TensorFlow and Keras. The model is trained on a dataset of facial expressions and can be converted to different formats for deployment. The notebook includes steps for data preparation, model training, evaluation, and inference.

Table of Contents

1. Overview
2. Dataset
3. Installation
4. Usage
5. Model Conversion
6. Inference
7. Results
8. License

Overview

The project aims to classify facial expressions into six emotion categories:

• Angry
• Happy
• Neutral
• Sad
• Surprise
• Ahegao

The model is built using a pre-trained MobileNetV2 as the base and fine-tuned for emotion recognition. It includes data augmentation, training, evaluation, and conversion to formats like TensorFlow Lite and TensorFlow.js for deployment.

Dataset

The dataset used for this project is the Emotion Recognition Dataset. It contains images of facial expressions categorized into different emotion classes.

Installation

1. Install the required Python packages:

   pip install -r requirements.txt

2. Download the dataset: The dataset is downloaded using the kagglehub library in the notebook.

Usage

1. Open the notebook ImageClassification.ipynb in your preferred IDE (e.g., PyCharm or Jupyter Notebook).

2. Follow the steps in the notebook:

   • Data preparation
   • Data augmentation
   • Model training
   • Evaluation and visualization
   • Model conversion
   • Inference

3. Run the cells sequentially to train the model and test its performance.

Model Conversion

The trained model is converted into the following formats for deployment:

• TensorFlow Lite: Saved in the tflite_model directory.
• TensorFlow.js: Saved in the tfjs_model directory.
• SavedModel: Saved in the saved_model directory.

Conversion Commands

• TensorFlow Lite:

  converter_tf_lite = tf.lite.TFLiteConverter.from_keras_model(model)
  tflite_model = converter_tf_lite.convert()
  with open('tflite_model/model.tflite', 'wb') as f:
      f.write(tflite_model)

• TensorFlow.js:

  tensorflowjs_converter --input_format=keras model.h5 tfjs_model

Inference

The notebook includes a sample inference pipeline using the TensorFlow Lite model. To test the model:

1. Provide the path to a test image.
2. Run the predict_image_tflite function to get the predicted class and probabilities.

Example:

test_image_path = 'data/final_dataset/test/Angry/sample_image.png'
predicted_class, predictions = predict_image_tflite(test_image_path)
print(f"Predicted class: {predicted_class}")

Results

The model achieved high accuracy on the test set. Below is an example of the output for a test image:

• Predicted Class: Angry
• Predicted Probabilities:

  Ahegao: 0.0000
  Angry: 0.9998
  Happy: 0.0000
  Neutral: 0.0001
  Sad: 0.0000
  Surprise: 0.0001