• Installations
• Project Motivation
• File Descriptions
• Using the Command Line
• Results
• Acknowledgements

All of the following packages were used for this project and can be found in the standard Anaconda distribution for Python 3.7:

• NumPy
• Matplotlib
• PIL
• Pytorch (torch and torchvision)

The flower image data set can be found here.

As part of Udacity's Data Scientist Nanodegree program, I created an image classifier based on a data set of 102 types of flowers. The classifier consists of a deep convolutional neural net that was pretrained on ImageNet images and available through PyTorch's torchvision.models package. For this project, I had to load and preprocess the image set, train the classifier on the image set, and then predict flower types for a given image. These steps can be run either in a Jupyter notebook, or from the command line of a terminal.

Files 1-3 are required to run the Jupyter notebook. Files 2-7 are required to run via command line.

1. Image Classifier Project.ipynb: Jupyter notebook that walks through all of the steps of preprocessing, training, and prediction.
2. workspace_utils.py: Contains functions provided by Udacity for keeping a workspace active while training on a GPU
3. cat_to_name.json: JSON file mapping the labels to flower names for the flower image set
4. train.py: Contains the general workflow for training an ImageNet model, testing its accuracy, and saving a model checkpoint. Requires inputs from the command line.
5. predict.py: Contains the general workflow for predicting flower types for a given image based on a trained model. Requires inputs from the command line.
6. model_functions.py: Creates a model class that contains the architecture of the classifier. Also includes the functions to train, test, and predict.
7. helper_functions.py: Various functions to prepare for model training or prediction: argument parsers from the command line, loading and saving checkpoints, processing and unprocessing images.

Train a new network on a data set with train.py

• Basic usage:

  python train.py data_directory

• Prints out training loss, validation loss, and validation accuracy as the network trains

• The data_directory must have folders within it labeled train, valid, and test and within each of those must be folders with index labels (1, 2, 3, etc) containing images

• Options:

  • Set directory to save checkpoints (default checkpoint.pth):

    python train.py data_dir --save_dir save_directory

  • Choose architecture (alexnet or vgg19_bn):

    python train.py data_dir --arch "alexnet"

  • Set hyperparameters (defaults: 0.001, 1000, 3):

    python train.py data_dir --learning_rate 0.01 --hidden_units 512 --epochs 20

  • Use GPU for training (default off):

    python train.py data_dir --gpu

Predict flower name from an image with predict.py along with the probability of that name. You'll pass in a single image /path/to/image and return the flower name and class probability.

• Basic usage:

  python predict.py /path/to/image checkpoint

• Options:

  • Return top K most likely classes (default 5):

    python predict.py input checkpoint --top_k 3

  • Use a mapping of categories to real names (default cat_to_name.json):

    python predict.py input checkpoint --category_names cat_to_name.json

  • Use GPU for inference (default off):

    python predict.py input checkpoint --gpu

Running train.py will print out the epoch number, training loss, validation loss, and validation accuracy after every 10 batches of 32 images.

Running predict.py will output the image provided and the top K categories along with their probabilities.

Stack Overflow posts and the documentation for each of the python packages were extremely helpful in completing this project.