Deep Neural Network Image Classifier

This repository contains some of the files I used for my Artificial Intelligence Programming With Python Nanodegree at Udacity.

The project consisted in training a neural network on a image dataset and predicting the content of new images based on what it learned. During the development stage the data was visualized using Matplotlib, Anaconda and Jupyter Notebooks.

The main tools used were

Library	Website
PyTorch	http://pytorch.org
NumPy	http://www.numpy.org
Matplotlib	https://matplotlib.org

Usage

Install the dependencies all the dependencies and run:

Training

Specify the path to the directory with the images.

$ python train.py data_directory_path

Predicting

Specify the path to the image to be predicted and to the saved model.

$ python predict.py image_file_path checkpoint_path

Data visualization

Example code used during development and testing:

def visualize(img_path, predictions, labels, label_mappings):
    true_label = img_path.split('/')[-2]
    predicted_range = np.arange(len(predictions))
    fig, (img_plot, label_plot) = plt.subplots(figsize=(12, 4), ncols=2, nrows=1)

    img_plot.set_xticks([])
    img_plot.set_yticks([])
    img_plot.set_title(label_mappings[true_label].capitalize())
    img_plot.imshow(Image.open(img_path))

    label_plot.set_yticks(predicted_range)
    label_plot.set_yticklabels([label_mappings[label] for label in labels])
    label_plot.invert_yaxis()
    label_plot.barh(predicted_range, predictions)

The following lines:

predictions, labels = predict('flowers/test/78/image_01848.jpg', loaded_model)
visualize('flowers/test/78/image_01848.jpg', predictions, labels, cat_to_name)

Will output the following results on a properly trained model: