Photo by Tara Winstead from Pexels

To be able to try the code you will need to

1. Download the flowers.zip file
2. Extract it to the root of the project as flowers/
3. Install the dependencies Using Anaconda (recommended)

conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
conda install -c anaconda pillow 

The flowers.zip has 3 folders that contain our flower images

• train this is our training data
• valid this is the data used for evaluating our classifier accuracy during training
• test used for sanity checking

The cat_to_name.json has the mappings between the flower ids and their actual names

Now it is time to do fun stuff!

• --arch: (optional) 'Set the CCN Model architecture to use'

  • vgg16
  • alexnet (default)

• --save_dir: (optional) 'Set the folder that will be used to save the checkpoints' the default is checkpoints

• --learning_rate: (optional) 'Set the learning rate' the default is 0.001

• --hidden_units: (optional) 'Set the number of hidden units in the classifier hidden layer' the default is 1024

• --epochs: (optional) 'Set the number of training epochs' default is 1

• --gpu: (optional) 'Train the model on gpu' this requires that you have a CUDA supported GPU!

The train.py script requires you to:

• pass as a first argument the folder that 'contains' your images
• after that you can pass in any order the above arguments as well to fine tune the training process 😸

If you have CUDA compatible gpu

python train.py flowers --epochs=15 --gpu

Otherwise

python train.py flowers --epochs=15

This will start the training process for each epoch the tool will train the classifier and will evaluate the classifier accuracy.

When the training is completed the tool will save a checkpoint in checkpoints/alexnet_checkpoint.pth

We will need this for making predictions later!

Now it is time to use our classifier!

• --category_names: (optional) 'Path to the category names JSON, this is used to map category IDs to their labels' the default is cat_to_name.json

• --top_k: (optional) 'The number of top predictions to be displayed' the default is 5

• --hidden_units: (optional) 'Set the number of hidden units in the classifier hidden layer' the default is 1024

• --gpu: (optional) 'Make prediction using the gpu' this requires that you have a CUDA supported GPU!

The predict.py script requires you to:

• pass as a first argument the path to your image
• pass as a seccond argument the checkpoint path to be used
• after that you can pass in any order the above arguments as well to fine tune the prediction process 😸

If you have CUDA compatible gpu

python predict.py flowers/test/10/image_07090.jpg checkpoints/alexnet_checkpoint.pth --gpu

Otherwise

python predict.py flowers/test/10/image_07090.jpg checkpoints/alexnet_checkpoint.pth

This will start the prediction process and you will get a list of the top predictions for the flowers/test/10/image_07090.jpg !

You have trained an image classifier and used it to make predictions 👏 !!!