A.I.ducation Analytics

Team Members

• Ronak 40221814
• Vijendra 40221273
• Aayush 40272388

Project Aim

The aim of this project is to develop an efficient and accurate system for classifying facial expressions from given photos using Convolutional Neural Networks (CNN).

Important

• Make sure the root directory is set as APPLIEDAI to make sure all the relative paths are working corrently.

Overview of Project Files

Core Scripts

• data_visualisation.py:

  • plot_histogram: Visualizes image distribution across expression categories.
  • display_random_images: Shows random images grid from four categories.
  • plot_histograms_from_paths: Plots histograms of pixel intensities.

• convert_greyscale_and_resize.py:

  • Automates conversion of images to grayscale and resizes them to 48x48 pixels.

• image_sharpening.py:

  • preprocess_images_in_directories: Enhances images with grayscale conversion, blurring, and unsharp masking.

• image_labeling.py:

  • Labels images in the format <class_label_name>_XXXX.jpg.

• filtering_locating_faces.py:

  • Detects and extracts faces from images.

• experiment.py:

  • Contains experimental code used during project development.

Model Training and Application

• train.py (Part-2 folder):

  • Trains the model using preprocessed and labeled images.

• application.py (Part-2 folder):

  • Loads the trained model for facial expression prediction.

Steps to Run the Project

Ensure the Data folder is set up correctly and all paths are relative to the Applied AI main directory.

1. Run filtering_locating_faces.py.
2. Run convert_greyscale_and_resize.py.
3. Run image_sharpening.py.
4. Run image_labeling.py.

Part-2: Model Creation and Testing

Prerequisites

• A folder named Data with four subfolders for facial expression classes.

Data Preparation

• Run data_prep.py.
• The dataset will be in dataset_inshape under the Part_2 directory.

Model Training

• Navigate to Part-2.
• Run train.py and choose option 2 for Model_variant2.
• The trained model is saved in Part_2/saved_model.

Running the Application

• Run application.py.
• Provide an image path for facial expression identification.
• Use external images for testing.

Part-3: Finding Bias in the model and Mitigating for it

This part of the project focuses on identifying any biases in the trained model and implementing strategies to mitigate these biases.

Prerequisites

• Dataset related folders such as DataKF, dataset_inshape, updated_datasetinshape, UpdatedDataKF with four subfolders for facial expression classes and saved_models .

Improvements over Step-2

• generate_confusion_matrix_on_test_set.py file is used to run an existing model over test dataset to generate confusion matrix and check the performance of the model
• helper_function.py file is modified to halt the training of the model if the accuracy keeps decreasing and saved the best model with the highest accuracy over all the successful execution
• train.py file is modified to use the same split of dataset to train and validate different models to maintain consistency in the results

Model Training

• Navigate to Part_3.
• Run train.py.
• The trained model will be saved in saved_models named best_model.pth.

Running the Application

• Run application.py.
• Provide an image path for facial expression identification.
• Use external images for testing.

Conclusion

Follow these instructions for successful project execution. For support or contributions, open an issue or submit a pull request.