lgalkina / flowers-image-classifier

Classification of flower images

Flowers-Image-Classifier

This project is part of Udacity's AI Programming with Python Nanodegree.

The goals of this project is to classify flower images using a pretrained CNN model.

Versions

• Anaconda 5.3 with Python 3
• PyTorch 0.4.1

Command Line Application

To train a new network on a data set:

• Run: python train.py
• The current epoch, training loss, validation loss, and validation accuracy as the netowrk trains will be printed
• Command line arguments:
  • Image folder as --dir with default value './flowers'
  • CNN model architecture as --arch with default value 'vgg'
  • Learning rate as --learning_rate with default value 0.001
  • Hidden units as --hidden_units with default value [512], pass several times to add multiple hidden units
  • Dropout as --dropout with default value 0.5
  • Number of epochs as --epochs with default value 20
  • Path to save checkpoint to as --checkpoint_path with default value './checkpoint.pth'
  • Processing unit --process_unit with default value 'gpu'

To predict flower name from an image:

• Run: python predict.py
• Command line arguments:
  • Path to image to predict as --image with default value './flowers/test/1/image_06764.jpg'
  • Path to retrieve checkpoint to as --checkpoint_path with default value './checkpoint.pth'
  • Top K predictions as --topk with default value 5
  • Path to JSON file containing label names as --labels with default value './cat_to_name.json'
  • Processing unit --process_unit with default value 'gpu'