AI Programming - Image Classification Neural Network

Project code for Udacity's AI Programming with Python Nanodegree program.

In this project we train an image classifier to recognize the different species of flowers. The first part of this project can be found in the notebook Image Classifier Project.ipynb. Here is where we work through defining the neural network architecture, process our images and train/validate and test our network's accuracy.

Note, an HTML version of this notebook can be found here: 'Image Classifier Project.html'

Along with the notebook this classification network can be ran on the command line.

For the dataset we are using, you can set those up with the following:

mkdir -p data_images && cd data_images
curl https://s3.amazonaws.com/content.udacity-data.com/nd089/flower_data.tar.gz | tar xz

1. Train

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 Options:

Set directory to save checkpoints: python train.py data_dir --save_dir save_directory

Choose architecture: python train.py data_dir --arch "vgg13"

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

Use GPU for training: python train.py data_dir --gpu

usage: python train.py ./data_sets/train --gpu --learning_rate 0.02 --epohcs 10 --arch vgg11

Train a network on a dataset

positional arguments:
  data_dir              Directory of the training images

optional arguments:
  -h, --help            show this help message and exit
  --gpu                 Use GPU for training, defaults to False
  --arch ARCH           Choose an architecture for the network, defaults to VGG19
  --save_dir SAVE_DIR   Set directory to save checkpoints
  --learning_rate LEARNING_RATE
                        Set the learning rate hyperparameter, defaults to 0.01
  --hidden_units HIDDEN_UNITS
                        Set the hidden unit amount, defaults to 512
  --epochs EPOCHS       Set the total amount of epochs this network should train for, defaults to 20

Available Architecture's:

arch_types = {
    'densenet121': models.densenet121,
    'densenet169': models.densenet169,
    'densenet161': models.densenet161,
    'densenet201': models.densenet201,
    'vgg11': models.vgg11,
    'vgg13': models.vgg13,
    'vgg16': models.vgg16,
    'vgg19': models.vgg19,
}

Example, training the network:

> python train.py ./data_images --arch "vgg19" --learning_rate 0.01 --gpu --epochs 10
---
Info -- Using GPU
Info -- (13) Epochs.  Learning Rate: 0.01.  Hidden Units: 2048.  NN Arch: vgg19
1/13 Epochs
Epoch: 1/13  Training Loss: 6.293  Validation Loss: 4.084  Validation Accuracy: 0.166
Epoch: 1/13  Training Loss: 4.017  Validation Loss: 3.147  Validation Accuracy: 0.290

Checkpoints

When training a neural network it will automatically save a checkpoint file in ./checkpoints/checkpoint-[ARCHNAME].pth. You can override this with the flag --save_dir /my/save/dir.

2. Prediction

Predict flower name from an image with predict.py along with the probability of that name. That is, 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 KKK most likely classes: python predict.py input checkpoint --top_k 3
Use a mapping of categories to real names: python predict.py input checkpoint --category_names cat_to_name.json
Use GPU for inference: python predict.py input checkpoint --gpu

usage: python predict.py ./data_sets/valid/1/img.jpg

Predict a flower name from an image along with the probability of that name.

positional arguments:
  image_path            Path to image to predict
  checkpoint            Checkpoint file location

optional arguments:
  -h, --help            show this help message and exit
  --gpu                 Use GPU for training, defaults to False
  --top_k TOP_K         Return top K most likely classes
  --category_names CATEGORY_NAMES
                        Mapping file of class id to real names, defaults to cat_to_name.json

Example of predicting with a supplied image:

>  python predict.py ./data_images/valid/71/image_04517.jpg ./checkpoints/checkpoint-vgg19.pth --gpu
--
Info -- Using GPU
Image resized to: (256, 191), from: (667, 500)
left: 16.0, upper: -16.5, right: 240.0, lower: 207.5
---------------Results-----------------
Prediction: GAZANIA(71) with accuracy 0.99999
Prediction: BLACK-EYED SUSAN(63) with accuracy 0.00001
Prediction: ENGLISH MARIGOLD(5) with accuracy 0.00000
---------------------------------------

coreyjs/aipnd-project

AI Programming - Image Classification Neural Network

1. Train

Checkpoints

2. Prediction