AI Programming with Python Project

Project code for Udacity's AI Programming with Python Nanodegree program. In this project, students first develop code for an image classifier built with PyTorch, then convert it into a command line application.

The image classifier aims to recognize different species of flowers that commonly occur in the United Kingdom. The model has been trained using the 102 Category Flower Dataset from the Visual Geometry Group at the University of Oxford.

As part of the overall application architecture, pre-trained deep learning models from the torchvision models subpackage were used (vgg16 or densenet121). The classifier portions of the models were replaced with the student's work using beginner knowledge of fully connected network architecture.

Project Status

The model was trained over 20 epochs with a testing accuracy of 92%.

A sample output of the classification model within a Jupyter Notebook is shown below. The top image shows the actual species and the bottom horizontal bar graph plots the model's predictions with associated probabilities.

Dependencies

• All code is written in Python 3
• torch
• torchvision
• PIL
• matplotlib
• numpy
• json

Network Architecture

I used the pre-trained densenet121 model and froze the features portion of the neural network that contains the convolution and pooling layers. I replaced the classifier portion, the fully connected layers, with the following code:

classifier = nn.Sequential(nn.Linear(1024, 512),
                           nn.ReLU(),
                           nn.Dropout(p = 0.5),
                           nn.Linear(512, 102),
                           nn.LogSoftmax(dim = 1))
model.classifier = classifier

Description of Files

Python script files:

filename	description
Image Classifier Project.ipynb	Jupyter Notebook of project code with explanatory text
get_input_args.py	Two functions to retrieve user command line inputs for training and prediction scripts
load_checkpoint.py	Function to load parameters of a trained network from checkpoint.pth file
load_data.py	Two functions, one that loads data for training and one that loads a single file to be fed into classifier model. Contains image transform code
predict.py	Calls the prediction_model.py function using user command line inputs from get_input_args.py
prediction_model.py	Function that feeds an image into classifier neural network and return prediction with associated probability
save_checkpoint.py	Function that saves parameters of a trained network for later use
train.py	Calls the train_model.py function using user command line inputs from get_input_args.py and generates checkpoint.pth output file using save_checkpoint.py function
train_model.py	Function that builds classifier neural network using PyTorch torchvision models subpackage and user defined layers. Training and validation functionality
workspace_utils.py	Keeps Udacity workspace active during training

Input files:

filename	description
cat_to_name.json	Dictionary of integers as keys and actual flower names as values

Output files:

filename	description
checkpoint.pth	PyTorch model checkpoint file, used to rebuild trained model

To-Do List

• Add model architecture section to README
• Continue training model, try to get higher accuracy

Credits

Student: Joshua Cayanan

Flower icon in readme header made by Freepik from www.flaticon.com
Deep Learning icon in readme header made by Becris from www.flaticon.com

Copyright

This project is licensed under the terms of the MIT license and protected by the Udacity Honor Code and Community Code of Conduct.