Classifying Images using a Neural Network

This program classifies images from the CIFAR-10 dataset using a Convolutional Neural Network (CNN). The images are preprocessed and normalized then used to train a CNN consisting of convolutional, max pooling, dropout, fully connected, and output layers.

In-depth analysis and examples can be found in dlnd_image_classification.ipynb.

Install

• Python 3.6
  • I recommend installing Anaconda as it is alreay set up for machine learning
  • If unfamiliar with the command line there are graphical installs for macOS, Windows, and Linux
• TensorFlow
• tqdm

Dataset

The dataset is broken into batches to prevent your machine from running out of memory. The CIFAR-10 dataset consists of 5 batches, named data_batch_1, data_batch_2, etc.. Each batch contains the labels and images that are one of the following:

• airplane
• automobile
• bird
• cat
• deer
• dog
• frog
• horse
• ship
• truck

Parameters

image_recognition.py

You can view different example of the CIFAR-10 dataset by changing the values of the batch_id and the sample_id:

• batch_id - id for a batch (1-5)
• sample_id - id for a image and label pair in the batch

batch_id = 1
sample_id = 5
helper.display_stats(cifar10_dataset_folder_path, batch_id, sample_id)

train_image_recognition.py

You can experiment with the CNN by altering:

• depth - Alter the depths of the CNN layers using common memory sizes
  • 64
  • 128
  • 256
  • ...
• epochs - number of training iterations
• batch_size - set to highest number your machine has memory for using common memory sizes
• keep_probability - probability of keeping node using dropout

Example Output

Command Line

python image_recognition.py

*You must press [enter] to continue after example image appears.

python train_image_recognition.py

License

The image_classification program is a public domain work, dedicated using CC0 1.0. I encourage you to use it, and enhance your understanding of CNNs and the deep learning concepts therein. :)